CN111504944A - Statistical monitoring method of citric acid fermentation liquefied clear liquid based on near infrared spectrum - Google Patents

Abstract

The invention relates to a statistical monitoring method of citric acid fermentation liquefied clear liquid based on near infrared spectrum, which comprises the steps of collecting and preprocessing input and output data, dividing a data set into two parts of marked data and unmarked data, establishing a semi-supervised PP L S model and calculating a monitoring index.

Description

Statistical monitoring method of citric acid fermentation liquefaction liquid based on near-infrared spectroscopy

technical field

The invention relates to the field of citric acid fermentation, in particular to a near-infrared spectrum-based statistical monitoring method for citric acid fermentation liquefaction liquid.

Background technique

The citric acid fermentation liquefaction liquid is the product of corn as the raw material: firstly, the corn is pulverized to obtain corn flour, and then diluted with water into corn steep liquor, the sugar water and process water are prepared to obtain feeding water, and NaOH is used to adjust, and the corn steep liquor and feeding water are mixed together. Mixing in the batching tank, adding amylase to get the material liquid, which enters the primary ejector, and then the material liquid is divided into two paths, one of which enters the seed mixing tank, and the other enters the secondary ejector. pH is used for solid-liquid separation. The mixed liquid is cooled by the cooler and then separated by the plate and frame, and the filter residue and clear liquid are obtained after the pressure ratio of the plate and frame. The liquefied supernatant is very important for the preparation of the fermentation medium, which directly determines the quality of the medium, which in turn affects the yield of the citric acid fermentation process. Therefore, monitoring the production process of citric acid fermentation liquefied serum is crucial to the whole citric acid fermentation process. Total sugar and total nitrogen are the key process parameters of liquefied clear liquid. Usually, manual sampling and offline analysis are used to obtain the values of these two parameters, and the operation state of liquefied clear liquid is determined by the method of upper and lower limit judgment.

Near-infrared spectroscopy collects a large amount of process information from the molecular vibration level. It has the characteristics of non-destructiveness, fast analysis and high efficiency. It has received extensive attention and has gradually become popular in the fields of agriculture, petroleum, medicine and the environment. Begin to introduce near-infrared spectrometer to detect key parameters. However, when building a calibration model for NIR spectroscopy, the sampling rate of process variables and quality variables is usually required to be consistent, while in the citric acid liquefaction serum process, NIR spectroscopy analysis is fast, and the offline sampling cycle of total sugar and total nitrogen is long. In the past, most of the methods used the method of downsampling rate. According to the sampling time of total sugar and total nitrogen, different spectral data were selected to make the sampling rate of the two consistent, and the model of total sugar and total nitrogen was established, but a large amount of near-term data was lost. Infrared spectral information will reduce the accuracy of the modeling. After the calibration model is determined, the total sugar and total nitrogen values are usually estimated according to the model, and then the upper and lower limit methods are used to monitor the liquefied clear liquid.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a statistical monitoring method of citric acid fermentation liquefied clear liquid based on near-infrared spectroscopy, which mainly solves the following two problems: near-infrared spectroscopy semi-supervised PPLS modeling of total sugar and total nitrogen method. The detection period of total sugar and total nitrogen is large and the lag time is long, while near-infrared spectroscopy can realize real-time online measurement. Therefore, in order to make full use of the near-infrared spectral data, it is necessary to use more near-infrared spectral samples to build the model, resulting in some near-infrared spectral samples without corresponding total sugar and total nitrogen samples, which brings difficulties to the establishment of the model. Different from previous modeling methods with downsampling rate, the semi-supervised PPLS method preserves the spectral information of unlabeled data, which can improve the accuracy of the model. A pre-warning method for citric acid fermentation liquefied serum based on NIR. In the long-term operation, the upper and lower limits of total sugar and total nitrogen can be obtained by the method of artificial experience. Therefore, the conventional method uses NIR as the detection equipment of total sugar and total nitrogen. When the values of total sugar and total nitrogen are obtained through the PPLS model , to judge whether these values are within their upper and lower limits, if not, it is considered that the liquefied serum is faulty. The present invention introduces the idea of statistical monitoring, replaces the monitoring method of upper and lower limits, and can provide a kind of pre-warning, thereby solving the problem of post-alarming. By solving the problems in the above two aspects, the present invention can judge the process operation state according to the statistical distribution law of key parameters of the liquefaction process, realize advance warning, and is of great significance for enterprises to improve production efficiency, reduce production costs, and obtain ideal product quality.

In order to solve the above-mentioned technical problems, the present invention provides a method for statistical monitoring of citric acid fermentation liquefied clear liquid based on near-infrared spectroscopy, including: input and output data collection and preprocessing, and dividing the data set into labeled data and unlabeled data Two parts, establishing the semi-supervised PPLS model and calculating the monitoring indicators.

In one embodiment, the specific steps are as follows:

Step 1: Spectral data collection and preprocessing

1) Sample collection: collect samples of different days and different batches of citric acid fermentation liquefied clear liquid under normal operating process and under abnormal conditions, and label the samples to obtain a sample set;

2) Near-infrared spectrum collection and sugar content measurement of citric acid fermentation liquefaction clear liquid: measure the sugar content of some sample sets, and measure the near-infrared spectrum of all samples with a near-infrared spectrometer;

3) Clear liquid spectrum preprocessing and band selection: Preprocess the spectrum with the method of multiple scattering correction, and select the wavelength band of the preprocessed spectrum to obtain the effective information data of the spectrum;

Step 2: Data Classification

4) Recombine the data in the modeling data set: divide the data in the modeling data into two groups, one group is labeled data, the other group is unlabeled data, and the two groups of data are of equal amount;

5) Division of modeling set and prediction set: Divide the sample set into two parts, one part is the modeling data set, which contains part of the labeled data and part of the unlabeled data, which are used to obtain the parameters required for monitoring indicators; the other part is the prediction set, which also contains a part of labeled data and a part of unlabeled data, which is used to test the monitoring effect of the model;

Step 3: Establish a semi-supervised probabilistic partial least squares model

6) Model establishment: semi-supervised PPLS modeling is performed on the rearranged data in step 5 to obtain relevant parameter values;

N<K为样本采样数，D_x，D_y分别为过程变量及质量变量的个数，半监督PPLS的模型可表示为：半监督PPLS的模型可表示为：Suppose the input and output variables are:

N<K is the number of samples, _Dx and Dy are the number of process variables and quality variables, respectively. The model of semi-supervised _PPLS can be expressed as: The model of semi-supervised PPLS can be expressed as:

x _n =Pt _n +μ _x +ξ _n

y _n =Ct _n +μ _y +ε _n

Divide the complete data set S into labeled data S ₁ and unlabeled data S ₂ , the processed S ₁ and S ₂ are data with the same number of samples, and the complete data set S can be divided into labeled data S ₁ and unlabeled data S 2 For data S ₂ , the processed S ₁ and S ₂ are data with the same number of samples;

Step 4: Establish monitoring indicators

7) Establish monitoring indicators: According to the estimated total sugar and total nitrogen, determine the normal distribution law of total sugar and total nitrogen under normal operation, and define the statistical distribution range of total sugar and total nitrogen under normal operation;

8) Substitute the near-infrared spectrum of the online detection into the calibration model established in step 6 to obtain the estimated values of total sugar and total nitrogen, and then judge whether the estimated value satisfies the normal normal distribution. If so, the process is running normally. If the threshold is exceeded, it is judged that the process operation is faulty.

In one of the embodiments, a joint interval probability partial least squares method is used to perform band selection on the preprocessed spectrum.

In one embodiment, the Fehling method is used to determine the sugar content of a portion of the sample set.

In one embodiment, the "6σ" criterion is used to define the statistical distribution range of total sugar and total nitrogen under normal operation.

In one of the embodiments, "the complete data set S is divided into labeled data S ₁ and unlabeled data S ₂ , and the processed S ₁ and S ₂ are both data with the same number of samples, and the complete data set S can be divided into For the labeled data S ₁ and the unlabeled data S ₂ , the processed S ₁ and S ₂ are data with the same number of samples”

Specifically, it can be expressed as:

S=S ₁ ∪ S ₂ ={(x _n ,y _n )|n=1,,,N}∪{x _k |k=N+1,,,K}

The complete likelihood function can be expressed as:

The parameters to be estimated are:

In one of the embodiments, the EM algorithm is used to obtain the parameters;

Parameter initialization:

Calculate E-step:

Compute M steps: Θ _k+1 =arg maxQ(Θ,Θ _k )

When |L(Θ _k+1 )-L(Θ _k )|>10 ^-5 , k=k+1, jump to the calculation step E.

Based on the same inventive concept, the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor, the processor implements any one of the above when executing the program. steps of the method described.

Based on the same inventive concept, the present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of any one of the methods.

Based on the same inventive concept, the present application also provides a processor for running a program, wherein the program executes any one of the methods when the program runs.

Beneficial effects of the present invention:

The process of liquefying the clear liquid of citric acid mainly adopts manual sampling to analyze the values of total sugar and total nitrogen, and then judges the operation status of the process according to the operation experience. waste and affect subsequent citric acid fermentation production. The present invention introduces near-infrared spectroscopy to detect the citric acid liquefaction process, thereby realizing statistical monitoring.

Compared with the conventional PLS modeling and PPLS modeling in the near infrared, the semi-supervised PPLS modeling method adopted in the present invention is adopted. This method takes into account the different acquisition periods of near-infrared spectra and total sugar and total nitrogen in the modeling process, and can utilize the information of near-infrared spectra as much as possible.

On the other hand, compared with the upper and lower limit monitoring method based on manual experience, the present invention introduces the monitoring idea of statistical process on the basis of near-infrared spectrum detection, no longer relies on manual experience, and can realize advance warning, replacing the upper and lower limits The post-event alarm of the monitoring method provides useful guidance for production.

Description of drawings

Fig. 1 is the schematic diagram of the univariate statistical monitoring total sugar based on semi-supervised PPLS in the statistical monitoring method of citric acid fermentation liquefaction supernatant based on near-infrared spectroscopy of the present invention.

2 is a schematic diagram of the univariate statistical monitoring of total nitrogen based on semi-supervised PPLS in the statistical monitoring method of citric acid fermentation liquefaction supernatant based on near-infrared spectroscopy of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

A near-infrared spectroscopy-based semi-supervised PPLS monitoring method for citric acid fermentation liquefaction supernatant, which is divided into four parts: input and output data acquisition and preprocessing, and dividing the data set into labeled data and unlabeled data. Part, establish a semi-supervised PPLS model, and calculate the monitoring indicators. Specific steps are as follows:

Step 1: Spectral data collection and preprocessing

1) Sample collection: Collect samples of citric acid fermentation liquefaction liquid from different batches on different dates and under normal operating processes and under abnormal conditions, and label the samples to obtain a sample set.

2) Near-infrared spectrum collection and sugar content determination of citric acid fermentation liquefied clear liquid: the sugar content of some sample sets was measured by Fehling method, and the near-infrared spectrum of all samples was measured by near-infrared spectrometer.

3) Clear liquid spectrum preprocessing and band selection: Preprocess the spectrum by the method of multiple scattering correction, and select the wavelength band of the preprocessed spectrum by the combined interval probability partial least square method to obtain the effective information data of the spectrum.

Step 2: Data Classification

4) Recombine the data in the modeling data set: Divide the data in the modeling data into two groups, one group is labeled data, the other group is unlabeled data, and the two groups of data are of equal amount.

5) Division of modeling set and prediction set: Divide the sample set into two parts, one part is the modeling data set, which contains part of the labeled data and part of the unlabeled data, which are used to obtain the parameters required for monitoring indicators; the other part is the prediction The set also contains a part of labeled data and a part of unlabeled data, which is used to test the monitoring effect of the model.

Step 3: Establish a semi-supervised probabilistic partial least squares model

6) Model establishment: conduct semi-supervised PPLS modeling on the rearranged data in step 5 to obtain relevant parameter values.

N<K为样本采样数，D_x，D_y分别为过程变量及质量变量的个数，半监督PPLS的模型可表示为：半监督PPLS的模型可表示为：Suppose the input and output variables are:

N<K is the number of samples, _Dx and Dy are the number of process variables and quality variables, respectively. The model of semi-supervised _PPLS can be expressed as: The model of semi-supervised PPLS can be expressed as:

x _n =Pt _n +μ _x +ξ _n

y _n =Ct _n +μ _y +ε _n

Divide the complete data set S into labeled data S ₁ and unlabeled data S ₂ , the processed S ₁ and S ₂ are data with the same number of samples, and the complete data set S can be divided into labeled data S ₁ and unlabeled data S 2 For the data S ₂ , the processed S ₁ and S ₂ are data with the same number of samples, specifically, it can be expressed as:

S=S ₁ ∪ S ₂ ={(x _n ,y _n )|n=1,,,N}∪{x _k |k=N+1,,,K}

The complete likelihood function can be expressed as:

用EM算法求取参数。The parameters to be estimated are:

Use the EM algorithm to find the parameters.

Parameter initialization:

Calculate E-step:

Compute M steps: Θ _k+1 =arg maxQ(Θ,Θ _k )

When |L(Θ _k+1 )-L(Θ _k )|>10 ^-5 , k=k+1, jump to the calculation step E.

Step 4: Establish monitoring indicators

7) Establish monitoring indicators: According to the estimated total sugar and total nitrogen, determine the normal distribution law of total sugar and total nitrogen under normal operation, and define the statistical distribution of total sugar and total nitrogen under normal operation according to the "6σ" standard scope.

8) Substitute the near-infrared spectrum of the online detection into the calibration model established in step 6 to obtain the estimated values of total sugar and total nitrogen, and then judge whether the estimated value satisfies the normal normal distribution. If so, the process is running normally. If the threshold is exceeded, it is judged that the process operation is faulty.

The present invention introduces the near-infrared spectroscopy technology into the monitoring of the citric acid fermentation and liquefaction process, and based on the semi-supervised PPLS method, realizes the collection of production operation information from the level of molecular vibration, and simultaneously applies the statistical monitoring concept to process the estimated production information , to evaluate the operating status.

Normally, total sugar and total nitrogen conform to a normal distribution. This example first obtains its distribution by normal total sugar and total nitrogen. If the satisfied distribution of total sugar and total nitrogen is N(μ, Λ), in the statistical process monitoring, the state is judged by judging the Mahalanobis distance index between the total sugar and total nitrogen of the sample and the normal distribution, and the index is Index=( x-μ) ^T Λ ^-1 (x-μ). In univariate statistical process monitoring, for convenience, for new test samples, the "6σ" criterion can also be calculated, that is, when the confidence level is 99.99966%, the normal quality variable interval is (μ-6Λ ^1/2 , μ+6Λ ^1/2 ).

A specific application scenario of the present invention is introduced below:

In the figure, the first 15 are normal samples, and the last 15 are abnormal samples. The univariate statistical monitoring results of total sugar based on semi-supervised PPLS are shown in Figure 1. The 19th to 30th samples are abnormal samples that exceed the control limit, that is, 12 abnormal samples are detected and 3 abnormal samples are missed; The univariate statistical monitoring results based on semi-supervised PPLS are shown in Figure 2. The 17th, 18th, 22nd, 23rd, 29th, and 30th samples are abnormal samples that exceed the control limit, that is, 6 abnormal samples are detected and 9 are missed. Each abnormal sample is normal, and all normal samples are normal. Combining the monitoring results of total sugar and total nitrogen, the detection results of 15 normal samples are all correct, and 14 of 15 abnormal samples are detected, indicating that the monitoring method proposed in this patent can effectively judge the operation state of the process.

The statistical monitoring method of the citric acid fermentation liquefaction liquid based on near-infrared spectroscopy provided by the present invention has been described in detail above, and the following points need to be explained:

1. A statistical process monitoring method of citric acid liquefaction clear liquid is proposed, which realizes pre-warning, replaces the traditional post-alarming, and can give early warning before failure affects product quality.

2. Based on the first technical key point, near-infrared spectroscopy was introduced to detect the liquefied serum, and a near-infrared spectroscopy modeling method of semi-supervised PPLS was proposed, which can make full use of a large number of non-corresponding total sugar and total nitrogen analysis values. Spectral data, improving the accuracy of modeling.

3. Based on the first technical key point and the second technical key point, the statistical monitoring method and the near-infrared spectroscopy technology are combined to analyze the statistical distribution of total sugar and total nitrogen in the liquefaction process under normal conditions, and judge the near-infrared spectrum Whether the detected total sugar and total nitrogen meet the normal distribution, so as to achieve the purpose of monitoring the liquefaction process.

The above-mentioned embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims (10)

1. A statistical monitoring method for citric acid fermentation liquefied clear liquid based on near infrared spectrum is characterized by comprising the steps of collecting and preprocessing input and output data, dividing a data set into marked data and unmarked data, establishing a semi-supervised PP L S model and calculating monitoring indexes.

2. The statistical monitoring method of the citric acid fermentation liquefied clear liquid based on the near infrared spectrum as claimed in claim 1, characterized by comprising the following specific steps:

the method comprises the following steps: spectral data acquisition and preprocessing

1) Collecting samples: collecting samples of the citric acid fermentation liquefied clear liquid in different batches on different dates under normal operation process and abnormal conditions, and labeling the samples to obtain a sample set;

2) near infrared spectrum collection and sugar content determination of the citric acid fermentation liquefied clear liquid: measuring the sugar content of part of the sample set, and measuring the near infrared spectrum of all samples by using a near infrared spectrometer;

3) spectrum pretreatment and band selection of clear liquid: preprocessing the spectrum by using a multivariate scattering correction method, and selecting the waveband of the preprocessed spectrum to obtain effective information data of the spectrum;

step two: data classification

4) Recombining data in the modeled dataset: dividing data in the modeling data into two groups, wherein one group is labeled data, the other group is unlabeled data, and the two groups have equal data quantity;

5) dividing a modeling set and a prediction set: the method comprises the steps of dividing a sample set into two parts, wherein one part is a modeling data set and comprises a part of labeled data and a part of unlabeled data for obtaining parameters required by monitoring indexes, and the other part is a prediction set and also comprises a part of labeled data and a part of unlabeled data for testing the monitoring effect of a model;

step three: establishing a semi-supervised probability partial least square model

6) Establishing a model, namely performing semi-supervised PP L S modeling on the data rearranged in the step 5 to obtain relevant parameter values;

assume that the input variables and the output variables are:

N<k is the number of sample samples, D_x，D_yThe number of the process variables and the quality variables respectively, and the model of the semi-supervised PP L S can be expressed as:

x_n＝Pt_n+μ_x+ξ_n

y_n＝Ct_n+μ_y+_n

dividing the complete data set S into marking data S₁And unmarked data S₂S after treatment₁And S₂All of which are consistent sample number data, the complete data set S can be divided into marker data S₁And unmarked data S₂S after treatment₁And S₂All the data are data with consistent sample number;

step four: establishing a monitoring index

7) Establishing a monitoring index: determining a normal distribution rule of total sugar and total nitrogen under normal operation according to the estimated total sugar and total nitrogen, and defining a statistical distribution range of the total sugar and the total nitrogen under normal operation;

8) and (3) substituting the online detected near infrared spectrum into the correction model established in the step (6) to obtain estimated values of total sugar and total nitrogen, then judging whether the estimated values meet normal distribution, if so, judging that the process is normal in operation, and if so, judging that the process is failed in operation.

3. The method for statistical monitoring of citric acid fermentation liquefied serum based on near infrared spectrum according to claim 1, wherein the spectrum after pretreatment is band-selected by joint interval probability partial least squares.

4. The method for statistical monitoring of liquefied serum for citric acid fermentation based on near infrared spectroscopy as claimed in claim 1, wherein the sugar content of part of the sample set is measured by the fisher method.

5. The method for statistical monitoring of liquefied serum for citric acid fermentation based on near infrared spectroscopy as claimed in claim 1, wherein the "6 σ" criterion is applied in defining the statistical distribution range under normal operation of total sugar and total nitrogen.

6. The method of claim 1, wherein the complete data set S is divided into labeled data S₁And unmarked data S₂S after treatment₁And S₂All of which are consistent sample number data, the complete data set S can be divided into marker data S₁And unmarked data S₂S after treatment₁And S₂All data are consistent in sample number "

Specifically, it can be expressed as:

S＝S₁∪S₂＝{(x_n,y_n)|n＝1,,,N}∪{x_k|k＝N+1,,,K}

the complete likelihood function can be expressed as:

the parameters to be estimated are:

7. the statistical monitoring method of citric acid fermentation liquefied clear liquid based on near infrared spectrum according to claim 1, wherein the parameter is obtained by EM algorithm;

initializing parameters:

and E, calculating:

and (4) calculating M steps: theta_k+1＝argmaxQ(Θ,Θ_k)

When | L (theta)_k+1)-L(Θ_k)|＞10^-5And k is k +1, and the step E of calculation is skipped.

8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1 to 7 are implemented when the program is executed by the processor.

9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

10. A processor, characterized in that the processor is configured to run a program, wherein the program when running performs the method of any of claims 1 to 7.

Images