CN101216419A - Method for quickly detecting yellow wine quality index - Google Patents

Abstract

The invention relates to a detection method of quality index of yellow rice wine, in particular to a method for rapidly detecting quality index of yellow rice wine, which has simple operation, improved detection speed and greatly improved time efficiency and production efficiency. The method comprises the following steps of: (1) scanning a yellow rice wine sample to be detected by Fourier transform near infrared spectrometer and a liquid fiber probe to obtain the spectrum of the yellow rice wine sample; and (2) sending the spectrum of the yellow rice wine sample obtained in step (1) to a quantitative spectrum analysis software package, and calculating the contents of quality indexes of the yellow rice wine sample by using a pre-constructed quality index model of the yellow rice wine, wherein the quality indexes comprise one or more of alcohol degree, total acid, total sugar, ammonia nitrogen, non-sugar solids, volatile ester and color ratio. The invention has the advantages of simple operation, improved detection speed, no pollution in the detection process and nondestructive detection, and can be further used for production line to achieve online detection and control.

Description

A kind of method of fast detecting contained yellow wine quality index

Technical field

The present invention relates to the detection method of contained yellow wine quality index, refer in particular to a kind of method of fast detecting contained yellow wine quality index.

Background technology

Yellow rice wine is to be primary raw material with grains such as rice, milled glutinous broomcorn millet, corn, millet, wheats, through boiling, saccharification, fermentation, squeezing, filtration, the brewed wine of explained hereafter such as store, blend.Amino-acid nitrogen and alcoholic strength are two important physical and chemical indexs that influence the yellow rice wine quality in production process of yellow rice wine.Amino-acid nitrogen is also referred to as ammonia nitrogen or ammoniacal nitrogen, and it can reflect that yellow rice wine contains the level of total amino acid content.Alcoholic strength is meant in 100ml wine sample, alcoholic milliliter number, and it is related to the important content of enterprise product quality stability and quality examination, and these two indexs have realistic meaning to raising yellow rice wine quality, restriction preparation inferior rice wine production.Indexs such as total reducing sugar, non-sugared solid content, amino-acid nitrogen, volatilization fat and look rate also are the importances to the yellow rice wine quality examination in addition.

Detect the national standard of the indexs such as alcoholic strength, total acid, total reducing sugar, ammoniacal nitrogen, non-sugared solid content, volatilization ester and look rate in the yellow rice wine at present, still adopt conventional analytical approach, complex operation, complexity, resulting analysis result poor in timeliness.How it is measured fast and accurately and have great importance reducing cost, improving the quality of products.

Near infrared spectrum (NIRS) technology is a measuring technology that develops rapidly the eighties in 20th century, have quick, lossless, multicomponent is analyzed simultaneously, analytic process is pollution-free, the reproducibility of analysis results advantages of higher, has become the important tool in the modern analysis measuring technology.Enter nineties 21 century, along with the development that instrument, optical fiber and stoichiometry learn a skill, near-infrared spectral analysis technology is widely used in every field such as food, pharmacy, tobacco, agricultural, chemical industry, and has obtained remarkable economical and social benefit.Partial least square method (PLS) is the important content of Chemical Measurement.

Summary of the invention

For overcoming the above problems, the invention provides a kind of simple to operately, can accelerate test speed, improve the method for the fast detecting contained yellow wine quality index of ageing and production efficiency greatly.

The technical solution adopted in the present invention is: a kind of method of fast detecting contained yellow wine quality index may further comprise the steps:

1) with ft-nir spectrometer and liquid fiber probe yellow rice wine sample to be measured is scanned, obtain yellow rice wine sample spectrogram to be measured;

2) the yellow rice wine sample spectrogram to be measured that step 1) is obtained is called in the quantitative spectrochemical analysis software package, utilizes the contained yellow wine quality index model of setting up in advance, draws the content of the yellow rice wine sample index of quality to be measured.

Wherein, contained yellow wine quality index is one or more in alcoholic strength, total acid, total reducing sugar, ammoniacal nitrogen, non-sugared solid content, volatilization ester and the look rate.

Wherein, step 2) method of the foundation of described contained yellow wine quality index model and check optimization comprises the steps:

(1) collects yellow rice wine as the standard model collection of setting up model;

(2) with ft-nir spectrometer and liquid fiber probe standard yellow rice wine sample is scanned the near infrared light spectrogram of the collection that gets standard samples;

(3) set up model with the near infrared light spectrogram of the standard model collection of acquisition in (2), select modeling interval and preprocess method, utilize partial least square method to set up mathematical model according to the different index of quality;

(4) with reference to the chemical score of each index of quality of the yellow rice wine of each sample of detection method bioassay standard sample sets of GB GB/T13662-2000 and GB17946-2000 yellow rice wine;

(5), model is tested and optimize with internal chiasma proof method or quantitative spectrochemical analysis software package with reference to the chemical score in the step (4).

Further, the spectrum pre-service in the step (3) of described contained yellow wine quality index modelling and check optimization method is that one or more of first order derivative method, vector normalization method, minimum-maximum normalization method are united use.

The concrete grammar of the internal chiasma proof method in the step (4) of described contained yellow wine quality index modelling and check optimization method is: the each intersection rejected 1 or several sample, obtain the sample predicted value with the disallowable sample of other sample modeling and forecastings, carry out successively; And the coefficient of determination (R by comparative sample predicted value and its chemical score ²) and root-mean-square deviation (RMSECV) weigh the quality of model, wherein, RMSECV and R ²Calculate by following formula:

$\mathrm{RMSECV}=\sqrt{\frac{1}{M}\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}$ ${\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="S200710306842XE00011.png"\; state="\{\{state\}\}">R</figure-callout>}}^2=(1-\frac{\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}{\mathrm{\&Sigma;}{(y_i-y_m)}^2})$

Wherein: Differ _iRepresent the poor of the chemical score of i sample and sample predicted value, M is a sample number, y _iBe the chemical score of i sample, y _mMean value for M sample predicted value.

Further, the method that quantitative spectrochemical analysis software package in the step (4) of described contained yellow wine quality index modelling and check optimization method is tested to model is: the standard spectrum of incoming inspection sample sets and corresponding data, utilization is optimized the mathematical model of setting up the back check sample collection is predicted.Root-mean-square deviation (RMSEP) by comparison prediction and the root-mean-square deviation (RMSECV) of model crosscheck are weighed the performance of model.

Technique effect of the present invention is: on the basis of Chemical Measurement, set up model with partial least square method, adopt near-infrared spectral analysis technology, measure principal ingredients such as the alcoholic strength in the yellow rice wine sample, non-sugared solid content, total acid, total reducing sugar, ammoniacal nitrogen makes comparisons and carries out line style and return with existing state calibration method, set up correlation model, with the model of being set up the above-mentioned principal ingredient in the yellow rice wine sample of the unknown is predicted again.This method is simple to operate, can accelerate test speed, improves ageing greatly and production efficiency, and test process do not have chemical reaction, need not chemical reagent, and is pollution-free, belongs to Non-Destructive Testing, can further be applied on the production line, realizes online Detection ﹠ Controling.This detection method is quick and accurate, if can in yellow rice wine enterprise produces, apply, for yellow rice wine industry measure provide a cover efficiently, analytical approach accurately and rapidly, progressively promote the application of near infrared technology in the yellow rice wine field, be expected to become the new national standard method that detects the yellow rice wine principal ingredient.

Description of drawings

Fig. 1 is the near infrared light spectrogram of yellow rice wine

Fig. 2 is alcoholic strength predicted value and true value crosscheck figure

Fig. 3 is total acid predicted value and true value crosscheck figure

Fig. 4 is total reducing sugar predicted value and true value crosscheck figure

Fig. 5 is non-sugared solid content predicted value and true value crosscheck figure

Fig. 6 is amino-acid nitrogen predicted value and true value crosscheck figure

Fig. 7 is volatilization ester predicted value and true value crosscheck figure

Fig. 8 is look rate predicted value and true value crosscheck figure

Embodiment

1 materials and methods

1.1 instrument and equipment

Ft-nir spectrometer, semiconductor cool off highly sensitive InGaAs detecting device, select the liquid fiber probe for use, are reference with the air, and spectral resolution is 8cm ^-1, scanning times 32 times.

1.2 material

Setting up the used sample of model is to be selected from Shaoxing brewery (sample that is adopted when setting up prescheme is a same kind yellow rice wine of selecting different manufacturers and different batches for use), and every index has been selected 80 parts of yellow rice wine samples respectively for use.

1.3 method

1.3.1 the chemical assay of standard model collection content

With reference to the content of principal ingredient in the detection method mensuration yellow rice wine of GB GB/T13662-2000 and GB17946-2000 yellow rice wine, i.e. chemical score.Amino acid nitrogen content is changed to 0.1%-2.0%, the alcoholic strength content is 8.0%-20.0%, total acid range 3.0%-7.0%, and sugared content is 4.0%-53.0%, the solid content variation range is 20%-55%, and volatilization ester content variation range is 0.0%-0.5%.

1.3.2 the foundation of mathematical model and optimization

The present invention adopts partial least square method (PLS) as the chemometrics method of setting up mathematical model.And use the internal chiasma proof method that model is optimized and verifies.Promptly be each the intersection to reject one or several samples, with the disallowable sample of other sample modeling and forecastings, carry out successively, and pass through the coefficient of determination (R of comparative sample predicted value and chemical score ²) and root-mean-square deviation (RMSECV) weigh the quality of model.R ²Calculate by following formula with RMSECV:

$\mathrm{RMSECV}=\sqrt{\frac{1}{M}\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}$ ${\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="S200710306842XE00011.png"\; state="\{\{state\}\}">R</figure-callout>}}^2=(1-\frac{\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}{\mathrm{\&Sigma;}{(y_i-y_m)}^2})$

Wherein: Differ _iRepresent the poor of the chemical score of i sample and closs validation predicted value, M is a sample number, y _iBe the chemical score of i sample, y _mMean value for M sample intersection predicted value.

The also available external inspection set pair of the present invention model is tested: the standard spectrum of incoming inspection sample sets and chemical score, utilize the mathematical model of optimizing back foundation that the check sample collection is predicted, weigh the performance of model by the root-mean-square deviation (RMSEP) and the root-mean-square deviation (RMSECV) of model crosscheck of comparison prediction.

2 examples

2.1 the near infrared light spectrogram of yellow rice wine sample

Utilize ft-nir spectrometer and liquid fiber probe that the yellow rice wine sample is carried out scanning analysis, all obtained spectrogram (see figure 1) clearly.Distinct spectral absorption characteristics provides abundant information basis for the quantitative test of yellow rice wine.

2.2 yellow rice wine alcoholic strength model

2.2.1 alcoholic strength modelling

Adopt first order derivative that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 2, the predicted value of model and the coefficient R between the actual value ²Can reach 99.71, the cross validation root-mean-square deviation is 0.143, and the prediction mean deviation is low to moderate 0.0008%, and visible model prediction accuracy is very high.

2.2.2 alcoholic strength model testing

With the model of building up the alcoholic strength of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 1.The mean deviation that predicts the outcome is 0.85%, to predict the outcome and true value is carried out paired t-test, t value 0.6888 shows predict the outcome there was no significant difference as a result with chemical gauging of near infrared, and the detection that the alcoholic strength model of being set up is applied to the yellow rice wine alcoholic strength is accurately and reliably.

Table 1 predicted value and chemical method measured value comparative result

Tab.1 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						11.5 16.5 17.6 11.5	11.52 16.52 17.56 11.35	-0.022 -0.0247 0.0369 0.149	17.4 16.9 17.6 10	17.34 17.11 17.49 9.918	0.0596 -0.213 0.108 0.0822

17.4 11.3 17.4 11.4 17.6 12.5 16.3 9.7 17.2 11.4 16.7 11.4 11.4 17.2 16.6 17.3

17.51 11.41 17.34 11.45 17.5 12.54 16.46 9.831 17.14 11.54 16.73 11.21 11.25 17.02 16.69 17.32

-0.113 -0.113 0.0559 -0.0483 0.104 -0.0443 -0.16 -0.131 0.0554 -0.139 -0.0299 0.193 0.147 0.176 -0.0874 -0.01 86

14.7 11.6 17.3 17.1 17.8 17.5 11.6 11.8 17.4 17.6 17.5 17.6 9.6 17.6 16.9 17.5

14.57 11.47 17.3 17.01 17.71 17.66 11.65 11.52 17.42 17.37 17.7 17.55 9.885 17.56 16.88 17.51

0.13 0.129 -0.00291 0.0876 0.0899 -0.16 -0.0516 0.279 -0.0224 0.227 -0.201 0.0548 -0.285 0.0425 0.0209 -0.0145

2.2.3 alcoholic strength model prediction reappearance test

Get a certain yellow rice wine sample at random, this sample is carried out the scanning collection spectrogram, gather 10 spectrograms, by its alcoholic strength of model prediction with ft-nir spectrometer and liquid fiber probe.The relative standard deviation that calculating predicts the outcome (RSD) is to investigate the reappearance of this model.Predicting the outcome sees Table 2, and the RSD that alcoholic strength predicts the outcome is 0.7%, and reappearance is good, and the measuring accuracy of model can satisfy the requirement of analyzing.

Table 2. precision of forecasting model

Tab.2 The precision for models(n＝10)

Measure number of times	Alcoholic strength (%)
		1 2 3 4 5 6 7 8 9 10	16.31 15.94 16.10 16.09 16.10 16.26 15.99 16.00 16.21 16.03

Mean value standard deviation (SD) relative standard deviation (RSD)

16.103 0.1222 0.7％

2.3 the total acid profile of yellow rice wine

2.3.1 the foundation of total acid profile

Adopt first order derivative and vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 3, the predicted value of model and the coefficient R between the actual value ²Be 87.47, the cross validation root-mean-square deviation is 0.127, and the prediction mean deviation is low to moderate 0.0002%, and visible model prediction accuracy is higher.

2.3.2 total acid model testing

With the model of building up the total acid of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 3.The average relative standard deviation that predicts the outcome is 1.6%, to predict the outcome and true value is carried out paired t-test, t value 0.9712, show predict the outcome there was no significant difference as a result with chemical gauging of near infrared, show that the detection that total acid profile of being set up is applied to the yellow rice wine total acidity is accurately and reliably.

Table 3 predicted value and chemical method measured value comparative result

Tab.3 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						5.3 6 6 5 5.5 5.3 5.6 4.9 6.1 5.2 5.8 4.9	5.314 5.841 5.953 5.067 5.66 5.306 5.636 5.039 5.888 5.103 5.794 4.925	-0.0145 0.159 0.0473 -0.0667 -0.16 -0.00579 -0.0359 -0.139 0.212 0.0966 0.00603 -0.025	5.7 5.5 5.2 5.6 5.9 5.7 5.6 5.6 5.4 5 5.8 5.6	5.49 5.686 5.027 5.636 5.873 5.714 5.761 5.595 5.341 5.147 5.661 5.514	0.21 -0.186 0.173 -0.0356 0.0271 -0.0138 -0.161 0.00516 0.0589 -0.147 0.139 0.0857

5.4 5.2 5.3 5.8 6 5.8 5.7 5.8

5.215 5.189 5.201 5.697 6.285 5.834 5.743 5.779

0.185 0.011 0.0994 0.103 -0.285 -0.034 -0.0435 0.0207

5.6 5.5 5.2 5.1 5.6 5.6 5.5 5.5

5.644 5.63 5.017 5.134 5.669 5.634 5.554 5.618

-0.0442 -0.13 0.183 -0.0341 -0.0694 -0.0337 -0.0542 -0.118

2.4 yellow rice wine total reducing sugar model

2.4.1 the foundation of total reducing sugar model

Adopt vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 4, the predicted value of model and the coefficient R between the actual value ²Be 99.52, the cross validation root-mean-square deviation is 0.753, and the prediction mean deviation is low to moderate 0.04144%, and visible model prediction accuracy is higher.

2.4.2 total reducing sugar model testing

With the model of building up the total reducing sugar of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 3.The average relative standard deviation that predicts the outcome is 1.87%, to predict the outcome and true value is carried out paired t-test, t value 0.3417, show predict the outcome there was no significant difference as a result with chemical gauging of near infrared, show that the detection that the total reducing sugar model of being set up is applied to the yellow rice wine total reducing sugar is accurately and reliably.

Table 5 predicted value and chemical method measured value comparative result

Tab.5 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						51.5 28.8 25 51.5 21 51.7 23.6 50.7	52.3 28.95 24.49 52.58 21.33 50.32 24.72 50.38	-0.801 -0.155 0.511 -1.08 -0.328 1.38 -1.12 0.323	29 24.8 26.2 50.2 25.6 28.3 25.2 26.8	29.02 24.66 25.28 49.73 26.24 28.91 24.98 27.59	-0.0191 0.142 0.917 0.472 -0.637 -0.613 0.22 -0.793

20.6 39.6 29.4 49.5 25.6 50.2 17.2 51.5 51 30 25.2 26

21.05 38.39 29.3 48.42 25.74 48.7 17.21 51.06 50.75 31.29 25.84 26.36

-0.454 1.21 0.0972 1.08 -0.138 1.5 -0.0102 0.441 0.249 -1.29 -0.637 -0.365

23.8 47.9 50 25 26.6 24 24.6 23.6 23.8 47.9 24.8 27.3

23.43 48.74 51.28 25.53 27.36 23.88 24.08 24.95 24 49.57 24.41 27.25

0.367 -0.843 -1.28 -0.535 -0.759 0.121 0.522 -1.35 -0.2 -1.67 0.39 0.0535

2.5 the non-sugared solid content model of yellow rice wine

2.5.1 the foundation of non-sugared solid content model

Adopt first order derivative and vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 5, the predicted value of model and the coefficient R between the actual value ²Be 97.43, the cross validation root-mean-square deviation is 1.21, prediction mean deviation 0.012559%, and visible model prediction accuracy is higher.

2.5.2 non-sugared solid content model testing

With the model of building up the non-sugared solid content of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 7.The average relative standard deviation that predicts the outcome is 2.08%, to predict the outcome and true value is carried out paired t-test, t value 0.8762, show predict the outcome there was no significant difference as a result with chemical gauging of near infrared, show that the detection that the non-sugared solid content model set up is applied to the non-sugared solid content of yellow rice wine is accurately and reliably.

Table 7 predicted value and chemical method measured value comparative result

Tab.7 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						48.6	50.96	-2.36	41.6	40.96	0.643

53.6 39.9 52.5 37.8 48.7 42.4 53 38.3 42.6 52.7 45.1 40.7 49.9 28.2 49.5 52.5 40.7 44.7 51.6

54 41.48 53.57 39.06 49.44 40.17 50.43 38.02 41.34 51.2 44.58 41.03 50.22 27.3 48.31 50.7 40.54 44.62 54.25

-0.398 -1.58 -1.07 -1.26 -0.741 2.23 2.57 0.283 1.26 1.5 0.521 -0.325 -0.316 0.896 1.19 1.8 0.161 0.0808 -2.65

44 44.8 52.2 40.9 50.9 41.1 48.1 31.2 51.5 41 50.3 40.4 39.7 49.6 50.1 40.5 47.2 41 40.1

46.11 44.41 51.23 40.57 51.11 42.17 46.76 32.13 50.33 40.89 49.48 41.62 40.19 51.49 50.16 40.21 47.02 40.22 40.11

-2.11 0.391 0.97 0.332 -0.215 -1.07 1.34 -0.928 1.17 0.112 0.823 -1.22 -0.491 -1.89 -0.0593 0.293 0.1 8 0.778 -0.00543

2.6 yellow rice wine amino nitrogen model

2.6.1 the foundation of amino nitrogen model

Adopt first order derivative and vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 6, the predicted value of model and the coefficient R between the actual value ²Be 91.87, the cross validation root-mean-square deviation is 0.0216, and the prediction mean deviation is low to moderate 0.0009%.

2.6.2 amino nitrogen model testing

With the model of building up the amino nitrogen of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 9.The average relative standard deviation that predicts the outcome is 2.44%, to predict the outcome and true value is carried out paired t-test, t value 0.7918, show predict the outcome there was no significant difference as a result with chemical gauging of near infrared, show that the detection that the amino nitrogen model of being set up is applied to the yellow rice wine amino nitrogen is accurately and reliably.

Table 9 predicted value and chemical method measured value comparative result

Tab.9 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						0.71 0.8 0.87 0.8 0.71 0.85 0.67 0.79 0.67 0.78 0.55 0.85 0.67 0.8 0.7 0.71 0.8 0.84 0.82 0.78	0.7031 0.8318 0.8338 0.8149 0.6853 0.8348 0.6812 0.8056 0.6518 0.8223 0.5813 0.8459 0.683 0.8183 0.6831 0.6999 0.7886 0.8278 0.801 0.7795	0.0069 -0.0318 0.0362 -0.0149 0.0247 0.0152 -0.0112 -0.0156 0.0182 -0.0423 -0.0313 0.00413 -0.013 -0.0183 0.0169 0.0101 0.0114 0.0122 0.019 0.000545	0.78 0.57 0.83 0.8 0.83 0.81 0.65 0.69 0.71 0.83 0.75 0.78 0.8 0.7 0.71 0.84 0.79 0.79 0.8 0.62	0.8223 0.5496 0.807 0.8203 0.8236 0.8014 0.6393 0.7415 0.6886 0.8119 0.7772 0.787 0.794 0.7111 0.7064 0.8595 0.7605 0.7645 0.7965 0.6481	-0.0423 0.0204 0.023 -0.0203 0.00641 0.0086 0.0107 -0.0515 0.0214 0.0181 -0.0272 -0.00698 0.00604 -0.0111 0.00358 -0.0195 0.0295 0.0255 0.00346 -0.0281

2.7 yellow rice wine volatilization ester model

2.7.1 the foundation of volatilization ester model

Adopt first order derivative and vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 7, the predicted value of model and the coefficient R between the actual value ²Be 64.53, the cross validation root-mean-square deviation is 0.0217, and the prediction mean deviation is low to moderate 0.00075%.Because the amount of volatilization ester is considerably less, there is certain error near infrared when detecting micro constitutent, and the accuracy of this model of model is relatively poor.

2.7.2 volatilization ester model testing

With the model of building up the volatilization ester of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 11.The average relative standard deviation that predicts the outcome is 7.1%, to predict the outcome and true value is carried out paired t-test, t value 0.7531, show that there is not significant difference in the result that near infrared predicts the outcome with chemical gauging, show that the volatilization ester model of being set up can be applied to the detection of yellow rice wine volatilization ester, but there is certain error, still there are certain directive function in rice wine production and quality control.

Table 11 predicted value and chemical method measured value comparative result

Tab.11 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						0.34 0.2 0.2 0.27 0.24 0.26 0.35 0.24 0.32 0.34 0.26 0.27 0.32 0.28 0.25 0.27 03 0.34 0.25 0.28	0.2981 0.2312 0.2264 0.2491 0.2604 0.255 0.3543 0.2583 0.322 0.31 0.2624 0.3018 0.3294 0.3103 0.2193 0.2666 0.2772 0.3064 0.2391 0.3219	0.0419 -0.0312 -0.0264 0.0209 -0.0204 0.00505 -0.00431 -0.0183 -0.00197 0.03 -0.00243 -0.0318 -0.00935 -0.0303 0.0307 0.00343 0.0228 0.0336 0.0109 -0.0419	0.26 0.27 0.27 0.27 0.26 0.33 0.29 0.27 0.26 0.28 0.25 0.34 0.24 0.27 0.25 0.27 0.28 0.32 0.23 0.29	0.2561 0.31 0.2466 0.2642 0.2884 0.2979 0.2645 0.2483 0.2562 0.2858 0.267 0.3295 0.2546 0.2567 0.2541 0.2642 0.2513 0.2881 0.2584 0.3057	0.00393 -0.04 0.0234 0.00585 -0.0284 0.0321 0.0255 0.0217 0.00379 -0.00581 -0.017 0.0105 -0.0146 0.0133 -0.00413 0.00581 0.0287 0.0319 -0.0284 -0.0157

2.8 yellow rice wine look rate model

2.8.1 the foundation of look rate model

Adopt first order derivative and vector normalization that spectrum is carried out pre-service, utilize partial least square method to set up mathematical model.The gained model is carried out crosscheck, as Fig. 8, the predicted value of model and the coefficient R between the actual value ²Be 78.39, the cross validation root-mean-square deviation is 110, prediction mean deviation 1.18%.

2.8.2 look rate model testing

With the model of building up the look rate of yellow rice wine sample is predicted, predicted the outcome and national standard method measurement result and deviation thereof see Table 13.The average relative standard deviation that predicts the outcome is 4.8%, to predict the outcome and true value is carried out paired t-test, t value 0.6286 shows that there is not significant difference in the result that near infrared predicts the outcome with chemical gauging, shows that the look rate model of being set up can be applied to the detection of yellow rice wine look rate.

Table 13 predicted value and chemical method measured value comparative result

Tab.13 The compared result of predictive and actual data

True value	Predicted value	Deviation	True value	Predicted value	Deviation
						1310 2140 1850 1560 1650 1380 1860 1650 1690 2020 1280 1840 1550 1340 1330 1720 1980 1910 1640 1870	1462 1886 1916 1460 1730 1371 1966 1453 1766 1768 1157 1884 1609 1309 1428 1819 2019 1821 1720 1838	-152 254 -65.7 100 -79.6 8.92 -106 197 -75.6 252 123 -43.7 -58.7 31.2 -98.3 -98.5 -39 89.3 -80 31.8	1870 1820 1260 1750 2020 1850 1720 1750 1060 1620 1230 1900 1650 1780 1770 1370 1190 1800 1810 1710	1886 1839 1235 1852 1909 1829 1832 1698 1194 1589 1325 1821 1758 1806 1740 1376 1315 1808 1818 1786	-16.3 -19 24.7 -102 111 20.7 -112 52.5 -134 30.7 -95 78.9 -108 -25.9 29.7 -5.94 -125 -7.87 -8.44 -76.5

The detection of 3 yellow rice wine samples to be measured

1) with ft-nir spectrometer and liquid fiber probe yellow rice wine sample to be measured is scanned, obtain yellow rice wine sample spectrogram to be measured;

2) the yellow rice wine sample spectrogram to be measured that step 1) is obtained is called in the quantitative spectrochemical analysis software package, utilizes top foundation and checks the contained yellow wine quality index model of optimizing, and draws the content of the yellow rice wine sample index of quality to be measured.

The present invention has adopted near infrared spectroscopy, has analysis speed and soon, does not consume characteristics such as chemical reagent.The present invention can detect indexs such as alcoholic strength, total acid, total reducing sugar, ammoniacal nitrogen, non-sugared solid content, volatilization ester and look rate simultaneously also can detect single index, having overcome the shortcoming of conventional method of analysis, is a kind of fast detecting new method in yellow rice wine quality control and the detection.

Claims (6)

1. the method for a fast detecting contained yellow wine quality index may further comprise the steps:

1) with ft-nir spectrometer and liquid fiber probe yellow rice wine sample to be measured is scanned, obtain yellow rice wine sample spectrogram to be measured;

2) the yellow rice wine sample spectrogram to be measured that step 1) is obtained is called in the quantitative spectrochemical analysis software package, utilizes the contained yellow wine quality index model of setting up and check optimization in advance, draws the content of the yellow rice wine sample index of quality to be measured.

2. the method for a kind of fast detecting contained yellow wine quality index according to claim 1 is characterized in that: described contained yellow wine quality index is one or more in alcoholic strength, total acid, total reducing sugar, ammoniacal nitrogen, non-sugared solid content, volatilization ester and the look rate.

3. the method for a kind of fast detecting contained yellow wine quality index according to claim 1 and 2 is characterized in that: the method that the foundation of described contained yellow wine quality index model and check are optimized comprises the steps:

(1) collects yellow rice wine as the standard model collection of setting up model;

(2) with ft-nir spectrometer and liquid fiber probe standard yellow rice wine sample is scanned the near infrared light spectrogram of the collection that gets standard samples;

(3) set up model with the near infrared light spectrogram of the standard model collection of acquisition in (2), select modeling interval and preprocess method, utilize partial least square method to set up mathematical model according to the different index of quality;

(4) with reference to the chemical score of each index of quality of the yellow rice wine of each sample of detection method bioassay standard sample sets of GB GB/T13662-2000 and GB17946-2000 yellow rice wine;

(5), model is tested and optimize with internal chiasma proof method or quantitative spectrochemical analysis software package with reference to the chemical score in the step (4).

4. the method for a kind of fast detecting contained yellow wine quality index according to claim 3 is characterized in that: the preprocess method in the described step (3) is the first order derivative method, eliminate in constant offset, vector normalization method and the minimum-maximum normalization method one or more unites use.

5. the method for a kind of fast detecting contained yellow wine quality index according to claim 3, it is characterized in that: the concrete grammar of the internal chiasma proof method in the step (4) of described contained yellow wine quality index modelling and check optimization method is: the each intersection rejected one or several samples, obtain the sample predicted value with the disallowable sample of other sample modeling and forecastings, carry out successively; And the coefficient of determination (R by comparative sample predicted value and its chemical score ²) and root-mean-square deviation (RMSECV) weigh the quality of model, wherein, RMSECV and R ²Calculate by following formula:

$\mathrm{RMSECV}=\sqrt{\frac{1}{M}\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}$ $R^2=(1-\frac{\mathrm{\&Sigma;}{\left({\mathrm{Differ}}_i\right)}^2}{\mathrm{\&Sigma;}{(y_i-y_m)}^2})$

Wherein: Differ _iRepresent the poor of the chemical score of i sample and sample predicted value, M is a sample number, y _iBe the chemical score of i sample, y _mMean value for M sample predicted value.

6. the method for a kind of fast detecting contained yellow wine quality index according to claim 3, it is characterized in that: the method that the quantitative spectrochemical analysis software package in the described step (4) is tested to model is: the standard spectrum of incoming inspection sample sets and chemical score, utilize the mathematical model of optimizing back foundation that the check sample collection is predicted, weigh the performance of model by the root-mean-square deviation (RMSEP) and the root-mean-square deviation (RMSECV) of model crosscheck of comparison prediction.

Images