Correlation analysis method between corn starch process parameters and starch milk DE value

Abstract

The embodiment of the present invention provides a correlation analysis method between corn starch process parameters and starch milk DE value, which belongs to the field of process modeling of big data, including: using principal component analysis to perform feature extraction and dimensionality reduction processing on initial input data to obtain feature vectors of initial input data and principal component input data of initial input data; using a convolutional neural network model to train the principal component input data, and calculating the total weight matrix of the trained convolutional neural network model; using a matrix formed by sequentially combining the feature vectors of the initial input data as the weight matrix of the initial input data; multiplying the total weight matrix of the convolutional neural network model by the weight matrix of the initial input data as the correlation matrix corresponding to the corn starch process site and the starch milk product DE value. According to the correlation between the corn starch process site and the starch milk product DE value, the present invention can quickly form a process adjustment plan for different raw materials to improve the starch milk DE value.

Description

Technical Field

[0001] The present invention relates to the field of process modeling of big data, and in particular to a method for analyzing the correlation between corn starch process parameters and starch milk DE value. Background Art

[0002] Corn starch industry is the basic industry of carbohydrate derivatives, and corn starch in its products is an important raw material for food processing. In terms of national life, economy and agricultural industrialization development, the starch industry in food processing is an important part of my country's food industry and one of the most important links in corn deep processing.

[0003] Corn raw materials from different sources will have different appropriate processing parameters during processing. With the refinement of management, the data collected by corn deep processing enterprises during the production process is becoming increasingly complex, the amount of data is growing explosively, and the data are all time series data. Faced with a huge amount of data records, it is difficult for production managers to accurately analyze the correlation between data based solely on production experience. In the corn starch process, the DE value is an important indicator that reflects the quality of starch milk products. The DE value indicates the percentage of reducing sugar in the dry matter of the syrup; how to use historical process data to mine the process parameters that have the greatest impact on the DE value is a technical problem that we urgently need to solve. Summary of the invention

[0004] The purpose of the embodiment of the present invention is to provide a method for analyzing the correlation between corn starch process parameters and starch emulsion DE value, which mainly solves the correlation between corn starch process parameters and starch emulsion DE value to achieve a process adjustment plan to improve the starch emulsion DE value.

[0005] In order to achieve the above object, a correlation analysis method between corn starch process parameters and starch milk DE value is provided, comprising:

[0006] Acquire corn raw material data associated with a site in a corn starch process and raw production monitoring data in the corn starch process to form initial input data;

[0007] Using the principal component analysis method to perform feature extraction processing on the initial input data to obtain the feature vector of the initial input data; and using the principal component analysis method to perform dimensionality reduction processing on the initial input data to obtain the principal component input data of the initial input data;

[0008] The principal component input data is trained using a convolutional neural network model, and a total weight matrix of the trained convolutional neural network model is calculated; and a matrix formed by sequentially combining the eigenvectors of the initial input data is used as a weight matrix of the initial input data;

[0009] The total weight matrix of the convolutional neural network model is multiplied by the weight matrix of the initial input data, and the obtained matrix is ​​used as the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product in the corn starch process.

[0010] Preferably, the step of obtaining corn raw material data associated with a site in a corn starch process and raw production monitoring data in the corn starch process to form initial input data includes:

[0011] De-noising the corn raw material data associated with the sites in the corn starch process and the original production monitoring data in the corn starch process to obtain de-noised data;

[0012] The Lasso regression analysis method is used to perform feature selection on the denoised data, and the data after feature selection is used as the initial input data.

[0013] Preferably, the method of performing feature extraction processing on the initial input data using the principal component analysis method to obtain a feature vector of the initial input data includes:

[0014] The initial input data is subjected to a standardization transformation to obtain a standardization matrix Z, and the standardization matrix Z is used to obtain a correlation coefficient matrix R between the standardization matrix Z and a transposed matrix of the standardization matrix Z;

[0015] Using the characteristic equation of the correlation coefficient matrix R, we get the unit eigenvector The feature vector used as the initial input data.

[0016] Preferably, the use of principal component analysis to perform dimensionality reduction processing on the initial input data to obtain principal component input data of the initial input data includes:

[0017] The normalized matrix Z and the unit eigenvector The matrix obtained by using the correlation formula is used as the principal component input data of the initial input data; the correlation formula is:

[0018] j=1,2,...,m; m is the number of principal components.

[0019] Preferably, the matrix formed by sequentially combining the feature vectors of the initial input data as the weight matrix of the initial input data includes:

[0020] The unit eigenvector As column vectors in turn, the obtained matrix is ​​used as the weight matrix W of the initial input data p :

[0021] m is the number of principal components; p is the number of characteristic sites.

[0022] Preferably, the convolutional neural network model includes multiple one-dimensional convolution kernels and a fully connected neural network; the principal component input data includes a training set and a validation set;

[0023] The method of training the principal component input data using a convolutional neural network model includes:

[0024] The principal component input data is convolved twice using multiple one-dimensional convolution kernels in the training set, and the data is input into a fully connected neural network for iterative training until the mean square error of the training set and the validation set converges, and the training is completed.

[0025] Preferably, the calculating of the total weight matrix of the trained convolutional neural network model includes:

[0026] 1) Add the weights of the same feature at different times in the convolutional layer to get the weight w of the convolutional layer jc :

[0027]

[0028] Among them, s is the number of convolution kernels; p is the size of the convolution kernel; n is the size of the convolution kernel c after the convolution operation; wh is the weight coefficient in the convolution kernel; wq is the weight coefficient in the fully connected layer: wq = wq1wq2,...,wq a ; Where a is the number of fully connected layers;

[0029] 2) The weight w of the convolutional layer jc The combined matrix is ​​used as the total weight matrix W of the convolutional neural network model c :W c =(w 1c ,...,w jc ).

[0030] Preferably, the correlation matrix W corresponding to the corn starch process site and the DE value of the starch milk product is:

[0031] W=W p W c .

[0032] Preferably, the association analysis method further includes:

[0033] The absolute values ​​of the elements in the correlation matrix corresponding to the corn starch process sites and the DE values ​​of starch milk products are sorted to obtain a sorted sequence; and the influence of the corn starch process sites on the DE values ​​is determined according to the order of the sorted sequence.

[0034] Preferably, the degree of influence includes positive correlation and negative correlation;

[0035] When the element in the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product is negative, it is determined that the influence of the corn starch process site corresponding to the element on the DE value of the starch milk is negatively correlated;

[0036] When an element in the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product is positive, it is determined that the influence of the corn starch process site corresponding to the element on the DE value of the starch milk is positively correlated.

[0037] Through the above technical solutions, the principal component analysis method PCA reduces the dimension of the characteristic sites of corn starch process data, which solves the dimensionality disaster problem of the data to a certain extent; the convolutional neural network CNN model also captures the time characteristics to solve the problem of inconsistent time series of data, making the modeling of big data more accurate and reliable. Finally, the weighted connection method WCM can intuitively and easily understand the correlation of the characteristic sites that affect the DE value of starch milk products. Finally, according to the correlation, a process adjustment plan can be quickly formed for different raw materials to improve the DE value of starch milk products.

[0038] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0040] Figure 1 Schematic diagram of a correlation analysis method between corn starch process parameters and starch milk DE value provided in the embodiment;

[0041] Figure 2 Schematic diagram of the contribution rate and cumulative contribution rate of the principal component provided in the embodiment;

[0042] Figure 3 It is a schematic diagram of a convolutional neural network model provided in an embodiment;

[0043] Figure 4 Schematic diagram of the training results of the convolutional neural network model provided in the embodiment;

[0044] Figure 5 It is a detection diagram of the DE value of the starch milk product before and after adjusting the parameters corresponding to the sites determined by the present invention provided in the embodiment. DETAILED DESCRIPTION

[0045] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0046] Example 1

[0047] like Figure 1 As shown, the present embodiment provides a method for analyzing the correlation between corn starch process parameters and starch milk DE value, comprising:

[0048] Acquire corn raw material data associated with a site in a corn starch process and raw production monitoring data in the corn starch process to form initial input data;

[0049] The principal component analysis method is used to perform feature extraction processing on the initial input data to obtain the feature vector of the initial input data; and the principal component analysis method is used to perform dimensionality reduction processing on the initial input data to obtain the principal component input data of the initial input data; here, the initial input data is mainly subjected to feature extraction and dimensionality reduction;

[0050] The principal component input data is trained using a convolutional neural network model, and a total weight matrix of the trained convolutional neural network model is calculated; and a matrix formed by sequentially combining the feature vectors of the initial input data is used as a weight matrix of the initial input data; machine learning through the convolutional neural network model is mainly performed after secondary feature extraction through a convolution kernel;

[0051] The total weight matrix of the convolutional neural network is multiplied by the weight matrix of the initial input data, and the obtained matrix is ​​used as the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product. According to the correlation matrix, the influence degree can be determined, so that the characteristic site parameters with large influence can be used to optimize the production process parameters and improve the DE value of the starch milk product. Among them, the DE value reflects the quality of the starch milk product in the corn starch process.

[0052] Preferably, the corn raw material data associated with the sites in the corn starch process and the original data of production monitoring in the corn starch process are obtained to form the initial data; the initial data is matrix data, and the columns of the matrix data are the physical and chemical data in the corn raw material data and the processing technology data in the original data of production monitoring in the corn starch process, and the order of arrangement is not required; the behavior time (in time order, the interval is 10 minutes), and then the feature selection of the initial data is performed. Specifically, the corn raw material data associated with the sites in the corn starch process and the original data of production monitoring in the corn starch process are denoised to obtain denoised data;

[0053] The Lasso regression analysis method is used to perform feature selection on the denoised data, and the data after feature selection is used as the initial input data.

[0054] Specifically, the specific method of denoising can be one or more data processing methods of adjusting the residence time, removing outliers, supplementing missing values, removing noise and reducing feature dimensions. The specific methods of various data processing methods are shown in the following steps 1.1)-1.4):

[0055] 1.1) Considering that the same batch of materials will have a long residence time in different processes, simply intercepting the process time recording site at the same time will cause material batch dislocation, and using the "present" input parameters to associate the target parameters affected by the "past" input. Therefore, based on the original corn data, the post-process data is associated with the residence time: the starch process data and starch detection data time is 48h (corn soaking), the saccharification stage time is 50h (residence in the liquefaction column), and the fructose OCC data time is 98h (saccharification time). In this way, it is ensured that each column of data in the time row of the input data matrix corresponds to the same batch of corn raw materials, ensuring the relative consistency of data information.

[0056] 1.2) Due to the stability of the process equipment sensor, the process data recorded may have abnormal values. Identify the abnormal values ​​that appear in the process and delete all the site data corresponding to the time. For example, if the data of column j (site W) of the data matrix is ​​abnormal at the i-th time (T moment), delete the i-th row in the data matrix. The determination of abnormal values ​​preferably uses the Chauvelet method to identify the abnormal values ​​of the data; the judgment formula of the Chauvelet method is:

[0057]

[0058] ω n =1+0.4ln(n) (2)

[0059] in, is the mean of the data, S x is the standard deviation of the data, ω n is the Chauveille coefficient, n is the number of data points; using formula (1), we can determine the outliers in the data.

[0060] 1.3) Since the time interval of corn raw material data is greater than the set value, it is different from the data of subsequent corn starch process operation, so the corn raw material data needs to be filled to ensure data stability. The interpolation method is used to fill the missing data, as shown in formula (3-4):

[0061]

[0062] x t_i =xt_n +k*(in) (4)

[0063] Among them, x t_n and x t_m is the known data before and after the missing value; K is the newly added missing data.

[0064] 1.4) Due to the limitations of process operation stability and detection technology, the collected data will contain a lot of noise, which greatly affects the model training results. Therefore, it is necessary to perform noise reduction on the original process data. After the outlier processing of the data matrix, this process does not need to judge the site data, and all data in the data matrix are denoised by column. The "(2n+1 point) simple moving average" method is used to smooth the filter to reduce the noise of the data, as shown in formula (5):

[0065]

[0066] Among them, x′ i is the denoised data, x i+1 For x i is the data of n before and after the center.

[0067] The above-mentioned Lasso regression analysis method is used to perform feature selection on the denoised data in order to select meaningful features for the next step of model training. The definition function of the Lasso regression analysis method is shown in formula (6):

[0068]

[0069] Among them, λ is a non-negative positive parameter that controls the complexity of the model; β is the regression coefficient vector; n is the number of features. The larger the λ is, the greater the penalty for the linear model with more features, thus obtaining a model with fewer features. This model selects λ=0.01.

[0070] The data after feature selection is standardized to obtain a standardized matrix Z, and the standardized matrix Z is used to obtain a correlation coefficient matrix R between the standardized matrix Z and the transposed matrix of the standardized matrix Z; the specific standardized conversion formula (7) is:

[0071]

[0072] Among them, n is the number of time dimension data, p is the number of feature sites, is the average value of the characteristic site data, and s is the variance of the characteristic site data. ij represents the elements in the normalized matrix Z, where i and j are the corresponding row and column indices.

[0073] Using the characteristic equation of the correlation coefficient matrix R, we get the unit eigenvector As the eigenvector of the initial input data. Specifically, the characteristic equation of the correlation coefficient matrix is:

[0074] |R-λI p |=0 (8)

[0075] Solve to obtain p characteristic roots λ;

[0076] Use formula (9-10) to solve and obtain the unit eigenvector

[0077]

[0078] Rb=λ j b,j=1,2,...,m (10)

[0079] Among them, m is the number of principal components determined, and the contribution rate of the principal component K j and the cumulative contribution rate S m like Figure 2 shown.

[0080] Preferably, the use of principal component analysis to perform dimensionality reduction processing on the initial input data to obtain principal component input data of the initial input data includes:

[0081] The normalized matrix Z and the unit eigenvector The matrix obtained by using the correlation formula is used as the principal component input data of the initial input data; the correlation formula is:

[0082] j=1,2,...,m; m is the number of principal components; Z i is the row vector of the normalization matrix Z.

[0083] Preferably, the matrix formed by sequentially combining the feature vectors of the initial input data as the weight matrix of the initial input data includes:

[0084] The unit eigenvector As column vectors in turn, the obtained matrix is ​​used as the weight matrix W of the initial input data p :

[0085] m is the number of principal components; p is the number of characteristic sites.

[0086] Preferably, Figure 3 As shown, the convolutional neural network model includes multiple one-dimensional convolution kernels and a fully connected neural network; the principal component input data includes a training set and a validation set;

[0087] The method of training the principal component input data using a convolutional neural network model includes:

[0088] First, multiple one-dimensional convolution kernels are used to perform convolution operations on the principal component input data, and then the data is input into the fully connected layer neural network. The convolution and neural network are iteratively trained until the mean square error of the training set and the validation set converges. The calculation of the mean square error is shown in formula (12). Figure 4 As shown in the figure, the mean square errors of the training set and the validation set are both close to zero, so the model training is good and there is no overfitting.

[0089]

[0090] Among them, y i is the true value, y′ i is the predicted value.

[0091] The reason for using a one-dimensional convolution kernel is to capture the impact of time series on the corn starch process, while the relative position characteristics between site data are not helpful for the knowledge learning of this process data, otherwise the machine learning model will over-exploit the characteristics of the data. The role of training a fully connected neural network model and mapping the learned "distributed feature representation" to the sample label space is equivalent to linking the features extracted by the convolution operation with the target, and obtaining the implicit rules in the data through continuous iterative training and learning.

[0092] Preferably, the calculating of the total weight matrix of the trained convolutional neural network model includes:

[0093] 1) Add the weights of the same feature at different times in the convolutional layer to get the weight w of the convolutional layer jc :

[0094]

[0095] Among them, s is the number of convolution kernels; p is the size of the convolution kernel; n is the size of the convolution kernel c after the convolution operation; wh is the weight coefficient in the convolution kernel; wq is the weight coefficient in the fully connected layer: wq = wq1wq2,...,wq a ; Where a is the number of fully connected layers;

[0096] 2) The weight w of the convolutional layer jc The combined matrix is ​​used as the total weight matrix W of the convolutional neural network model c :W c =(w 1c ,...,w jc ).

[0097] Preferably, the correlation matrix W corresponding to the corn starch process site and the DE value of the starch milk product is: W = W p W c .

[0098] Preferably, the association analysis method further includes:

[0099] The absolute values ​​of the elements in the correlation matrix corresponding to the corn starch process sites and the DE values ​​of starch milk products are sorted to obtain a sorted sequence; and the influence of the corn starch process sites on the DE values ​​is determined according to the order of the sorted sequence.

[0100] Preferably, the degree of influence includes positive correlation and negative correlation;

[0101] When the element in the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product is negative, it is determined that the influence of the corn starch process site corresponding to the element on the DE value of the starch milk is negatively correlated;

[0102] When an element in the correlation matrix corresponding to the corn starch process site and the DE value of the starch milk product is positive, it is determined that the influence of the corn starch process site corresponding to the element on the DE value of the starch milk is positively correlated.

[0103] For the total weight matrix W c By sorting the absolute values ​​of the elements in the formula, we can obtain the characteristic sites that have a greater impact on the DE value target of starch milk products, which can provide certain guidance for the next step of process parameter tuning. The results are shown in Table 1.

[0104] Table 1: Correlation table between characteristic locus number and DE value

[0105] Characteristic sites Correlation with DE value LIA2103_7 Negative correlation LIA2103_3 Positive correlation LIA2103_5 Negative correlation LIA2103_8 Positive correlation LIA2103_1 Positive correlation PID1_LIC1401_2_5-PV Positive correlation LEVEL1_LIA_1473_1 Positive correlation LEVEL2_LIA_1590 Negative correlation PID1\LIC_302A-PV Negative correlation PID1\LIC_502A-PV Positive correlation PID1_LIC1401_2_12-PV Negative correlation PID1\LIC_501A-PV Negative correlation PID1\LIC_301B-PV Positive correlation PID1\LIC_502B-PV Negative correlation LEVEL1_LIA_1401_1_5 Positive correlation LEVEL1_LIA_1684 Positive correlation LEVEL2_LIA_1401_3_2 Negative correlation LEVEL1_LIA_1401_1_4 Negative correlation LIA2103_2 Positive correlation LEVEL1_LIA_1631 Positive correlation

[0106] Adjust the characteristic sites with greater influence to optimize the production process parameters. For example, the characteristic site parameters with greater positive and negative influence on the DE value of starch milk products are adjusted by 10% respectively. The results for a period of time are as follows: Figure 5 As shown, the DE value increased by 21% cumulatively.

[0107] From the above technical scheme, it can be seen that the present invention has the following advantages: by analyzing the corn starch raw material and process data, characteristic parameters that affect the DE value of starch milk products for different corn raw materials are obtained, and the results have good universality. By adjusting the mined characteristic parameters, the DE value of the product is improved, providing a better auxiliary effect for corn starch processing and production.

[0108] Through the above five steps of data processing, principal component analysis, convolutional neural network modeling, model weight extraction and determination of optimized characteristic site parameters, the correlation analysis between corn starch process parameters and starch milk product DE values ​​was carried out. Feature selection and dimensionality reduction were performed on corn starch production data, which not only avoided the dimensional disaster problem of big data modeling, but also ensured the independence of the input machine learning model data. At the same time, feature extraction and model training were performed twice, which made the results obtained by this technical solution more reliable while mining and extracting the time series features of process data. This technical solution can also be used for big data mining of other processes to adjust production process parameters. Thereby optimizing product production, improving product quality and maintaining enterprise production safety.

[0109] The optional implementation modes of the embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation modes. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical scheme of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention.

[0110] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0111] In addition, various implementation modes of the embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed by the embodiments of the present invention.

Claims

1. A method for analyzing the correlation between corn starch process parameters and starch milk DE value, characterized in that: include: Acquire corn raw material data associated with a site in a corn starch process and raw production monitoring data in the corn starch process to form initial input data; Using the principal component analysis method to perform feature extraction processing on the initial input data to obtain the feature vector of the initial input data; and using the principal component analysis method to perform dimensionality reduction processing on the initial input data to obtain the principal component input data of the initial input data; The principal component input data is trained using a convolutional neural network model, and a total weight matrix of the trained convolutional neural network model is calculated; and a matrix formed by sequentially combining the eigenvectors of the initial input data is used as a weight matrix of the initial input data; The total weight matrix of the convolutional neural network model is multiplied by the weight matrix of the initial input data, and the obtained matrix is ​​used as the correlation matrix corresponding to the corn starch process site and the starch milk DE value.

2. The analysis method according to claim 1, characterized in that The corn raw material data associated with the site in the corn starch process and the original production monitoring data in the corn starch process are obtained to form initial input data, including: De-noising the corn raw material data associated with the sites in the corn starch process and the original production monitoring data in the corn starch process to obtain de-noised data; The Lasso regression analysis method is used to perform feature selection on the denoised data, and the data after feature selection is used as the initial input data.

3. The analysis method according to claim 1 or 2, characterized in that The method of using the principal component analysis method to perform feature extraction processing on the initial input data to obtain the feature vector of the initial input data includes: The initial input data is subjected to a standardization transformation to obtain a standardization matrix Z, and the standardization matrix Z is used to obtain a correlation coefficient matrix R between the standardization matrix Z and a transposed matrix of the standardization matrix Z; Using the characteristic equation of the correlation coefficient matrix R, we get the unit eigenvector The feature vector used as the initial input data.

4. The analysis method according to claim 3, characterized in that The principal component analysis method is used to perform dimensionality reduction processing on the initial input data to obtain the principal component input data of the initial input data, including: The normalized matrix Z and the unit eigenvector The matrix obtained by using the correlation formula is used as the principal component input data of the initial input data; the correlation formula is: m is the number of principal components.

5. The analysis method according to claim 3, characterized in that The matrix formed by sequentially combining the feature vectors of the initial input data as the weight matrix of the initial input data includes: The unit eigenvector As column vectors in turn, the obtained matrix is ​​used as the weight matrix W of the initial input data p : m is the number of principal components; p is the number of characteristic sites.

6. The analysis method according to claim 1, characterized in that The convolutional neural network model includes multiple one-dimensional convolution kernels and a fully connected neural network; the principal component input data includes a training set and a validation set; The method of training the principal component input data using a convolutional neural network model includes: The principal component input data is convolved twice using multiple one-dimensional convolution kernels in the training set, and the data is input into a fully connected neural network for iterative training until the mean square error of the training set and the validation set converges, and the training is completed.

7. The analysis method according to claim 6, characterized in that The method of calculating the total weight matrix of the trained convolutional neural network model includes: 1) Add the weights of the same feature at different times in the convolutional layer to get the weight w of the convolutional layer jc : Among them, s is the number of convolution kernels; p is the size of the convolution kernel; n is the size of the convolution kernel c after the convolution operation; wh is the weight coefficient in the convolution kernel; wq is the weight coefficient in the fully connected layer: wq = wq1wq2,...,wq a ; Where a is the number of fully connected layers; 2) The weight w of the convolutional layer jc The combined matrix is ​​used as the total weight matrix W of the convolutional neural network model c :W c =(w 1c ,...,w jc ).

8. The analysis method according to claim 7, characterized in that The correlation matrix W corresponding to the corn starch process site and the DE value of starch milk product is: W=W p IN c 。 9. The analysis method according to claim 1, characterized in that The analysis method further comprises: The absolute values ​​of the elements in the correlation matrix corresponding to the corn starch process sites and the starch milk DE values ​​are sorted to obtain a sorted sequence; and the influence of the corn starch process sites on the starch milk DE value is determined according to the order of the sorted sequence.

10. The analysis method according to claim 9, characterized in that The degree of influence includes positive correlation and negative correlation; When the element in the correlation matrix corresponding to the corn starch process site and the starch milk DE value is negative, it is determined that the influence of the corn starch process site corresponding to the element on the starch milk DE value is negatively correlated; When an element in the correlation matrix corresponding to the corn starch process site and the DE value of starch milk is positive, it is determined that the influence degree of the corn starch process site corresponding to the element on the DE value of starch milk is positively correlated.

Images