Composite material starch content inversion model construction method based on response surface method

Abstract

The invention discloses a response surface method-based composite material starch content inversion model construction method, and relates to the technical field of material processing. The response surface method-based composite material starch content inversion model construction method comprises the following steps of: analyzing a distribution state and a distribution rule of starch content in a composite material by utilizing a spectral measurement technology, and determining main factors influencing the starch content in the composite material; combining a response surface method with a matrix combination algorithm to determine a main factor combination; constructing a coupling model based on a main factor combination and a constraint function; and establishing an inversion model based on the coupling influence degree, and performing inversion prediction on the starch content in the composite material by using the inversion model. The inversion model provided by the invention can process the inversion of the starch content in the complex composite material, has relatively strong applicability, and can provide a global view angle for the overall starch content distribution, so that the change trend of different regions or stages of the composite material can be found.

Description

Technical Field

[0001] The invention relates to the technical field of material processing, and in particular to a method for constructing an inversion model of starch content in a composite material based on a response surface methodology. Background Art

[0002] Composite materials are made up of two or more different materials. Better performance and characteristics can be obtained through their combination. Starch, as a filler, can be filled in the matrix of the composite material to change the density and volume of the composite material. The addition of starch can increase the mass of the composite material and improve the strength and rigidity of the composite material. At the same time, starch acts as a reinforcing agent in the composite material to increase the mechanical strength and rigidity of the composite material. By controlling the content and dispersibility of starch, the mechanical properties of the composite material, such as bending strength, tensile strength and impact resistance, can be adjusted.

[0003] The determination of starch content plays an important role in the performance and characteristics of composite materials. In the process of determining the starch content in composite materials, an inversion model is usually established to perform inversion identification. The inversion model can accurately predict and estimate the starch content in the composite material, which is very important for material quality control and quality assessment.

[0004] In the prior art, spectral measurement technology is used to obtain a wealth of characteristic parameters related to starch content, and these characteristic parameters are used to analyze the main factors affecting the starch content in composite materials. However, since the factors affecting the starch content in composite materials are diverse and complex, if the combination of factors that have a significant impact on the starch content cannot be accurately screened out, it is impossible to evaluate and analyze the main factors and their interactions, and further impossible to systematically explore the interaction effects between multiple factors, resulting in the establishment of an inversion model containing a large number of minor factors, which increases the computational complexity and data redundancy and reduces the prediction accuracy of the inversion.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a method for constructing an inversion model of starch content in a composite material based on response surface methodology to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows: A method for constructing an inversion model of starch content in a composite material based on a response surface methodology, the method comprising the following steps: S1. Analyze the distribution state and distribution law of starch content in the composite material by using spectral measurement technology, and determine the main factors affecting the starch content in the composite material; S2, combining the response surface methodology with the matrix combination algorithm to determine the main factor combination; S3. Construct a coupling model based on the main factor combination and the constraint function to obtain the coupling influence of the main factor combination on the performance of the composite material; S4. An inversion model is established based on the degree of coupling influence, and the inversion model is used to perform inversion prediction on the starch content in the composite material.

[0008] Preferably, analyzing the distribution state and distribution law of the starch content in the composite material by using spectral measurement technology, and determining the main factors affecting the starch content in the composite material comprises the following steps: S11, collecting original near infrared spectrum data of the composite material using a near infrared spectrometer; S12, using a spectral feature selection algorithm to extract effective features from the original near-infrared spectral data, and dividing the composite materials into different clustering methods according to the classification index; S13, respectively screening spectral feature intervals for different clustering methods, and performing set operations on the spectral feature intervals to generate a joint matrix; S14, using an orthogonal linear discriminant analysis algorithm to transform the original near-infrared spectral data after feature extraction into an orthogonal linear discriminant analysis space, and generate a projection matrix; S15, performing similarity measurement calculation on the projection matrix, and generating an optimal spectral feature subset based on the similarity measurement calculation result; S16. Analyze the distribution state and distribution law of starch content in the composite material based on the optimal feature subset, and extract spectral feature representation therefrom; S17. Analyze the main factors affecting starch content in the composite material based on the spectral characteristics.

[0009] Preferably, the original near-infrared spectral data after feature extraction is transformed into an orthogonal linear discriminant analysis space using an orthogonal linear discriminant analysis algorithm, and generating a projection matrix includes the following steps: S141, adding historical near infrared spectrum data of the same type of composite materials to the original near infrared spectrum data after feature extraction to obtain fused near infrared spectrum data; S142, performing partial least square feature extraction on the fused near-infrared spectrum data to obtain a least square feature matrix; S143, performing orthogonal linear discriminant feature extraction on the least square feature matrix in the orthogonal linear discriminant analysis space to obtain an orthogonal linear discriminant feature matrix; S144, converting the orthogonal linear discriminant feature matrix into a solution optimization matrix, sorting the solution optimization matrix, and forming a projection matrix with the feature vectors at the front of the sorting column.

[0010] Preferably, performing similarity measurement calculation on the projection matrix and generating the optimal spectral feature subset based on the similarity measurement calculation result comprises the following steps: S151, extracting feature points from the fused near-infrared spectrum data, calculating the distances between the feature points, and connecting each feature point with its nearest adjacent feature point to generate an adjacent undirected graph; S152, assigning a weight to each feature point in the adjacent undirected graph according to the distance between the feature points, and transforming the objective function by combining the weight with the Laplace matrix; S153, solving the target value based on the converted target function, obtaining the target vector corresponding to each feature point, and combining the target vectors to form a final projection matrix; S154, solving the final projection matrix to obtain a low-dimensional embedding matrix, and using the Euclidean distance to measure the similarity between the feature points in the low-dimensional embedding matrix space, and then constructing the optimal spectral feature subset based on the similarity between the feature points.

[0011] Preferably, combining the response surface method with the matrix combination algorithm to determine the main factor combination comprises the following steps: S21, respectively formulate the fitting relationship between starch content and each main factor, and generate combination categories using a matrix combination algorithm, and then calculate the dynamic change data of starch content in each combination category; S22, constructing a response surface model based on the dynamic change data, and using the response surface model to construct a polynomial response function between starch content and main factors; S23, using a data optimization iterative algorithm and Latin hypercube sampling to search for a minimum value in the polynomial response function, and using the minimum value as a parameter value; S24, generating a matrix classification combination between starch content and each main factor, and extracting the main factor combination from the matrix classification combination using a global optimization algorithm.

[0012] Preferably, respectively formulating the fitting relationship between starch content and each main factor, generating combination categories using a matrix combination algorithm, and then calculating the dynamic change data of starch content in each combination category comprises the following steps: S211, calculating the fitting relationship and preference degree between starch content and each main factor, and generating a combination category of starch content and each main factor; S212, analyzing the semantic relationship in the combined category space and the connection between the hierarchical structures, and calculating the semantic distance in the combined category space; S213, constructing a distance feature matrix model and a combined feature matrix model based on a combined category space based on semantic distance; S214, integrating the generated feature matrix model and the combined feature matrix model into the matrix decomposition model to obtain a matrix combination model based on the combined category space; S215. Calculate the dynamic change data of starch content in each combination category using a matrix combination model.

[0013] Preferably, analyzing the semantic relationship in the combined category space and the connection between the hierarchical structures and calculating the semantic distance in the combined category space comprises the following steps: S2121. Calculate the jump span between each combination category and the hierarchical structure in the Hasse diagram of the combination category space; S2122: Determine the preference distance between the combination categories according to the jump span, and calculate the semantic distance between two adjacent combination categories that jump forward and backward according to the preference distance.

[0014] Preferably, the expression of the distance feature matrix model is: ; In the formula, Indicates u The distance feature matrix of the main factors; represents the latent feature matrix of the u-th main factor; G represents the combined category space vector; T Indicates the total number of jumps; t Indicates t Jump times; A represents the indicator feature matrix; W The semantic distance between two adjacent combination categories representing the forward and backward jumps.

[0015] Preferably, establishing an inversion model based on the degree of coupling influence, and using the inversion model to perform inversion prediction on the starch content in the composite material comprises the following steps: S41. Calculate the first-order derivative and the second-order derivative based on the coupling influence degree, and construct the main factor parameters and the original near-infrared spectrum parameters; S42, performing correlation analysis on the original near infrared spectrum parameters and the main factor parameters, and determining the sensitive parameters for inverting starch content by using a cross-validation method; S43, constructing an inversion model based on sensitive parameters, and constraining relative weights in the inversion model according to prior information; S44. Based on the constraint condition of relative weight ratio, the inversion model is used to perform inversion prediction on the starch content in the composite material.

[0016] Preferably, constructing an inversion model based on sensitive parameters and constraining relative weights in the inversion model according to prior information comprises the following steps: S431, constructing an inversion model according to the sensitive parameters of the inverted starch content, and determining the constraint endpoint values ​​of the inversion model, and determining the prior information of the main factor parameters and the original near infrared spectrum parameters; S432, adding constraints to the relative weight ratio of the inversion model according to the prior information, and establishing a constrained composite model; S433, using a global optimization algorithm to determine the initial parameter values ​​of the inversion model, and setting a solution search method, and updating the relative weight value and parameter update value of the inversion model in the constrained composite model; S434, taking the relative weight value and the parameter update value as the constraint function of the inversion model, when the difference between adjacent constraint functions is less than a preset threshold, the iterative process is stopped, otherwise, steps S433 to S434 are repeated; S435, continuously updating the relative weight ratio and the parameter update value to obtain the final relative weight ratio and the final parameter update value.

[0017] The beneficial effects of the present invention are: 1. The present invention can obtain the absorption, reflection or transmission information of the composite material sample at different wavelengths through spectral measurement technology, providing rich multivariate information, which is helpful to fully understand the distribution state of the starch content in the composite material, and judge the main factors affecting the starch content from the distribution state, which is helpful to deeply understand the law of starch content distribution and the mechanism of action of influencing factors, and then provide data basis for the subsequent inversion identification of starch content in the composite material.

[0018] 2. The present invention can help systematically generate various factor combinations through the matrix combination algorithm, and can efficiently screen out the main factors that have a significant impact on the target variable through the response surface method, which helps to improve operational efficiency. At the same time, combined with the model establishment of the response surface method, a more detailed evaluation of the main factors and their interactions can be performed, which helps to deeply understand the degree of influence and mode of action of each factor on the target variable, and then sufficient information can be obtained to determine the main factor combination, reducing operating costs and time.

[0019] 3. The present invention establishes an inversion model through the coupling influence degree between the main factor combination on the composite material performance, so as to comprehensively consider the complex influence of each main factor on the starch content, and more comprehensively explore the interaction effect between different factors. By considering these interactions, the accuracy of the inversion model is improved, thereby avoiding the inversion model from being interfered by secondary factors, and improving the calculation efficiency and prediction accuracy of the inversion model.

[0020] 4. The inversion model provided by the present invention can handle the inversion of starch content in complex composite materials, has strong applicability, and can provide a global perspective on the overall starch content distribution, which is helpful to discover the changing trends of different regions or stages of the composite materials. At the same time, the inversion model can accurately predict and estimate the starch content in the composite material, which helps to adjust the ratio of raw materials and processing parameters so as to accurately obtain the required starch content, thereby improving production efficiency and production quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 The present invention is a flowchart of a method for constructing an inversion model of starch content in a composite material based on the response surface methodology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0024] According to an embodiment of the present invention, a method for constructing an inversion model of starch content in a composite material based on response surface methodology is provided.

[0025] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the method for constructing an inversion model of starch content in a composite material based on response surface methodology according to an embodiment of the present invention, the method for constructing an inversion model of starch content in a composite material based on response surface methodology comprises the following steps: S1. Use spectral measurement technology to analyze the distribution state and distribution law of starch content in composite materials, and determine the main factors affecting starch content in composite materials.

[0026] Among them, using spectral measurement technology to analyze the distribution state and distribution law of starch content in the composite material and determine the main factors affecting the starch content in the composite material includes the following steps: S11. Collect original near-infrared spectrum data of the composite material using a near-infrared spectrometer.

[0027] It should be noted that the use of a near-infrared spectrometer to collect raw near-infrared spectral data of a composite material includes the following steps: before spectral collection, the near-infrared spectrometer needs to be calibrated to ensure the precision and accuracy of the instrument; the composite material sample is scanned using a near-infrared spectrometer to record spectral data within a wavelength range, and the collected data will include the absorbance or reflectance information of the sample at different wavelengths; the collected raw spectral data is preprocessed, including denoising, baseline correction and other steps to improve the quality and analyzability of the data, and the processed spectral data is stored for subsequent analysis and modeling.

[0028] S12. Use spectral feature selection algorithm to extract effective features from the original near-infrared spectral data, and divide the composite materials into different clustering methods according to classification indicators.

[0029] It should be noted that the use of spectral feature selection algorithm to extract effective features from the original near-infrared spectral data and to divide the composite materials into different clustering methods according to the classification indicators includes the following steps: calculating the correlation coefficient between each wavelength point and the target variable in the original near-infrared spectral data; selecting wavelength points that are highly correlated with the target variable as features; using the variance analysis algorithm to evaluate the significant differences between samples of different categories at each wavelength point, and selecting wavelength points with significant differences as features; based on the concept of information theory, calculating the information gain of each wavelength point for classification, and selecting wavelength points with larger gains as features; using the PCA method to reduce the dimensionality of the original spectral data and extract the main components as features.

[0030] S13, respectively screening spectral feature intervals for different clustering methods, and performing set operations on the spectral feature intervals to generate a joint matrix.

[0031] S14. Using an orthogonal linear discriminant analysis algorithm, the original near-infrared spectral data after feature extraction is transformed into an orthogonal linear discriminant analysis space, and a projection matrix is ​​generated.

[0032] Among them, the original near-infrared spectral data after feature extraction is transformed into the orthogonal linear discriminant analysis space using the orthogonal linear discriminant analysis algorithm, and the projection matrix is ​​generated, including the following steps: S141. Add historical near-infrared spectral data of the same type of composite materials to the original near-infrared spectral data after feature extraction to obtain fused near-infrared spectral data.

[0033] S142. Perform partial least squares feature extraction on the fused near-infrared spectral data to obtain a least squares feature matrix.

[0034] It should be noted that the partial least squares feature extraction of the fused near-infrared spectral data to obtain the least squares feature matrix includes the following steps: using the partial least squares regression algorithm to establish a regression model between the fused near-infrared spectral data and the target variable; in the PLS regression model, each spectral wavelength point has a correlation coefficient, which indicates its importance to the target variable. According to these correlation coefficients, the spectral wavelength point that can best explain the target variable is selected as the least squares feature; the selected least squares features are arranged in the order of the wavelength points to construct a least squares feature matrix.

[0035] S143. Perform orthogonal linear discriminant feature extraction on the least square feature matrix in the orthogonal linear discriminant analysis space to obtain an orthogonal linear discriminant feature matrix.

[0036] S144, converting the orthogonal linear discriminant feature matrix into a solution optimization matrix, sorting the solution optimization matrix, and forming a projection matrix with the feature vectors at the front of the sorting column.

[0037] It should be noted that converting the orthogonal linear discriminant feature matrix into a solution optimization matrix, sorting the solution optimization matrix, and forming a projection matrix with the eigenvectors in the front column of the sorting include the following steps: using the orthogonal linear discriminant analysis (OLDA) algorithm to perform dimensionality reduction and feature extraction on the orthogonal linear discriminant feature matrix. In OLDA, the eigenvectors are combined into a feature matrix, and each column represents an eigenvector; solving the optimization matrix refers to performing certain calculations and transformations on the feature matrix to further extract useful features; sorting the eigenvectors in the solution optimization matrix according to their contribution to sample classification. The sorting can be based on the size of the eigenvalue. The larger the eigenvalue, the greater the contribution of the eigenvector to sample classification; forming a projection matrix with the eigenvectors in the front column of the sorting.

[0038] S15, performing similarity measurement calculation on the projection matrix, and generating an optimal spectral feature subset based on the similarity measurement calculation result.

[0039] The similarity measurement calculation is performed on the projection matrix, and the optimal spectral feature subset is generated based on the similarity measurement calculation result, which includes the following steps: S151, extracting feature points from the fused near-infrared spectral data, calculating the distances between the feature points, and connecting each feature point with its nearest adjacent feature point to generate an adjacent undirected graph.

[0040] It should be noted that extracting feature points from fused near-infrared spectral data, calculating the distance between feature points, and connecting each feature point with its nearest adjacent feature point to generate an adjacent undirected graph includes the following steps: extracting feature points from fused near-infrared spectral data using a window algorithm; calculating the distance between the extracted feature points, and the distance calculation can be based on the position information of the feature points, the similarity between the spectral data, etc.; connecting each feature point with its nearest adjacent feature point according to the distance between the feature points to form an adjacent undirected graph, and the connection can be represented by an edge, and the weight of the edge can be the distance between the feature points.

[0041] S152, assigning a weight to each feature point in the adjacent undirected graph according to the distance between the feature points, and transforming the objective function by combining the weight with the Laplace matrix.

[0042] It should be noted that assigning weights to each feature point in adjacent undirected graphs according to the distance between the feature points, and transforming the objective function by combining the weights with the Laplace matrix includes the following steps: assigning weights to each feature point in the adjacent undirected graphs according to the distance between the feature points, and the weights can be assigned according to the size of the distance; constructing a Laplace matrix based on the adjacent undirected graphs and the assigned weights. The Laplace matrix is ​​a matrix commonly used in graph theory to describe the structure and characteristics of the graph. The Laplace matrix can be divided into a symmetric Laplace matrix and an asymmetric Laplace matrix; transforming the objective function into a form combined with the Laplace matrix, replacing the feature point weight terms in the objective function with the diagonal elements of the Laplace matrix, so that the relationship between the feature points can be introduced into the objective function, so that the optimization process can more fully consider the structure and similarity between the feature points.

[0043] S153. Solve the target value based on the transformed target function to obtain the target vector corresponding to each feature point, and combine the target vectors to form a final projection matrix.

[0044] S154, solving the final projection matrix to obtain a low-dimensional embedding matrix, and using the Euclidean distance to measure the similarity between the feature points in the low-dimensional embedding matrix space, and then constructing the optimal spectral feature subset based on the similarity between the feature points.

[0045] S16. Analyze the distribution state and distribution law of starch content in the composite material based on the optimal feature subset, and extract spectral feature representation from it.

[0046] S17. Analyze the main factors affecting starch content in the composite material based on the spectral characteristics.

[0047] It should be noted that the main factors affecting the starch content in the composite material can be analyzed based on the spectral characteristics, which mainly reflect the following aspects: Spectral wavelength region: Spectral data usually contains multiple wavelengths. Changes in starch content may cause changes in the absorption peaks of the spectrum in different wavelength regions. Therefore, the main factors related to changes in starch content can be found out by analyzing the characteristics of spectral data in different wavelength regions.

[0048] Absorption peak intensity: For a specific wavelength region, the change in starch content can be determined by analyzing the intensity of the absorption peak in the spectral data. An increase in starch content may lead to an increase in the absorption peak intensity in a specific wavelength region, or a decrease in the absorption peak intensity in other wavelength regions.

[0049] Spectral shape: Changes in starch content may also affect the overall shape of the spectrum. The impact of changes in starch content on the spectral shape can be determined by analyzing the curve shape, peak and valley positions and other characteristics of the spectral data.

[0050] Spectral differences: By comparing the spectral data of samples with different starch contents, the spectral differences caused by changes in starch content can be found.

[0051] It should be noted that the main factors affecting the starch content in composite materials may also include other non-spectral characteristics, such as component ratio, processing technology, etc. Therefore, the relationship between spectral characteristics and other factors should be comprehensively considered during analysis to obtain more comprehensive and accurate analysis results.

[0052] S2. Combine the response surface methodology with the matrix combination algorithm to determine the main factor combination.

[0053] Among them, combining the response surface method with the matrix combination algorithm to determine the main factor combination includes the following steps: S21. The fitting relationship between starch content and each main factor is formulated respectively, and the combination category is generated by using the matrix combination algorithm, and then the dynamic change data of starch content in each combination category is calculated.

[0054] The following steps are included in the steps of respectively formulating the fitting relationship between starch content and each main factor, generating combination categories by using a matrix combination algorithm, and then calculating the dynamic change data of starch content in each combination category: S211. Calculate the fitting relationship and preference degree between starch content and each main factor, and generate combination categories of starch content and each main factor.

[0055] It should be noted that the calculation of the fitting relationship and preference degree between starch content and each main factor, and the generation of combination categories of starch content and each main factor include the following steps: using statistical regression analysis methods to take starch content as the dependent variable and each main factor as the independent variable to establish a regression model; using the index of model fitting degree to evaluate the explanatory power of each main factor on starch content; judging the preference degree of each main factor for starch content based on the coefficient of the regression model, the positive or negative value of the coefficient indicates the positive or negative influence of the factor on starch content, and the size of the coefficient indicates the intensity of the influence; clustering is performed according to the combination of starch content and each main factor, and the samples are divided into different combination categories according to the clustering results, and each category represents a combination of starch content and main factors.

[0056] S212: Analyze the semantic relationship in the combined category space and the connection between the hierarchical structures, and calculate the semantic distance in the combined category space.

[0057] The steps of analyzing the semantic relationship in the combined category space and the connection between the hierarchical structures and calculating the semantic distance in the combined category space include the following steps: S2121. Calculate the jump span between each combination category and the hierarchical structure in the Hasse diagram of the combination category space.

[0058] It should be noted that calculating the jump span between each combination category and the hierarchical structure in the Hasse diagram of the combination category space includes the following steps: constructing a Hasse diagram of the combination category space in the combination category space, the Hasse diagram is a directed acyclic graph, in which each combination category represents a node, and the directed edges between the nodes represent the inclusion relationship between the combination categories; determining the hierarchical structure of the combination categories, the hierarchical structure represents the nested relationship between the combination categories, in which the top layer is the most general combination category and the bottom layer is the most specific combination category; for each combination category, calculating the jump span between it and the hierarchy, the jump span refers to the shortest path length required to reach the corresponding node in the hierarchy from a combination category, and the jump span can be calculated by using the shortest path algorithm in graph theory.

[0059] S2122: Determine the preference distance between the combination categories according to the jump span, and calculate the semantic distance between two adjacent combination categories that jump forward and backward according to the preference distance.

[0060] It should be noted that judging the preference distance between combination categories according to the jump span, and calculating the semantic distance between two adjacent combination categories of the previous and next jumps according to the preference distance includes the following steps: the preference distance is to measure the distance between combination categories according to the jump span, a smaller jump span indicates that the relationship between the combination categories is closer, and a larger jump span indicates that the relationship between the combination categories is more distant; the jump span is normalized, for example, the jump span is divided by the maximum jump span, and the range of the preference distance is between 0 and 1; the semantic distance is to calculate the distance between adjacent combination categories according to the preference distance, and the preference distance can be mapped to the range of the semantic distance using linear interpolation; the preference distance between each combination category and the calculated semantic distance are analyzed.

[0061] S213. Construct a distance feature matrix model and a combination feature matrix model based on the combination category space based on semantic distance.

[0062] Among them, the expression of the distance feature matrix model is: ; In the formula, Indicates u The distance feature matrix of the main factors; represents the latent feature matrix of the u-th main factor; G represents the combined category space vector; T Indicates the total number of jumps; t Indicates t Jump times; A represents the indicator feature matrix; W The semantic distance between two adjacent combination categories representing the forward and backward jumps.

[0063] It should be noted that in the distance feature matrix model, the latent feature matrix and the indicator feature matrix are two important matrices used to represent the relationship between objects and features; the latent feature matrix is ​​used to represent the representation of objects in the latent feature space, where each latent feature vector corresponds to a latent feature; the indicator feature matrix is ​​used to indicate the presence or absence of features between objects, and each object can be represented by a set of indicator feature vectors, where each indicator feature vector corresponds to a feature, and the element value of the indicator feature vector is usually binary, indicating whether the object has the corresponding feature.

[0064] S214, integrating the generated feature matrix model and the combined feature matrix model into the matrix decomposition model to obtain a matrix combination model based on the combined category space.

[0065] S215. Calculate the dynamic change data of starch content in each combination category using a matrix combination model.

[0066] It should be noted that the dynamic change data of starch content in each combination category calculated using the matrix combination model can be obtained through analysis and calculation, and the change trend of starch content in each combination category can be displayed through visualization methods (such as line graphs, bar graphs, etc.) to further analyze and explain the results.

[0067] S22. A response surface model is constructed based on the dynamically changing data, and a polynomial response function between starch content and main factors is constructed using the response surface model.

[0068] It should be noted that constructing a response surface model based on dynamically changing data and using the response surface model to construct a polynomial response function between starch content and main factors includes the following steps: determining dynamically changing data as the response variable, and being able to select multiple main factors to construct a polynomial response function according to actual conditions; covering different value ranges of the response variable to ensure that the data points are evenly distributed and fully covered; establishing a polynomial response function between starch content and main factors through regression analysis, and the constructed response surface model can be used to predict the value of starch content under given main factor values.

[0069] S23. Use data optimization iterative algorithm and Latin hypercube sampling to search for the minimum value in the polynomial response function, and use the minimum value as the parameter value.

[0070] It should be noted that the use of data optimization iterative algorithm and Latin hypercube sampling to search for the minimum value in the polynomial response function and taking the minimum value as the parameter value includes the following steps: using the gradient descent method to search for the minimum value of the polynomial response function; determining the parameter range of the polynomial response function; using the gradient descent method for iterative optimization; in the optimization process, combining Latin hypercube sampling to improve the search efficiency. Latin hypercube sampling is a uniform sampling method that can generate a set of uniformly distributed sample points in the parameter space for evaluating the value of the response function. By using Latin hypercube sampling in the optimization iteration process, the parameter space can be searched more comprehensively and the optimization process can be accelerated.

[0071] S24, generating a matrix classification combination between starch content and each main factor, and extracting the main factor combination from the matrix classification combination using a global optimization algorithm.

[0072] It should be noted that the present invention constructs a response surface model using dynamically changing data and generates a polynomial response function between starch content and main factors, thereby quantifying the relationship between starch content and each main factor through the response surface method, so that the complex influence between each factor can be represented in the form of a polynomial. The establishment of the response surface model not only improves the description ability of the established inversion model for the complex interactive effects of the main factors, but also effectively eliminates the influence of minor factors; through matrix classification combination and global optimization algorithm, the best combination is extracted from the combination category of the main factors, so that through global optimal search, the most representative factor combination is selected, so that the inversion model can more comprehensively and systematically analyze the influence of the factor combination on the starch content.

[0073] The present invention combines technologies such as response surface methodology, matrix combination algorithm and optimization algorithm to enable the constructed inversion model to achieve a systematic analysis of the main factors and their interaction effects, eliminate interference from minor factors, ensure that the model has high computational efficiency and prediction accuracy, and thus can accurately and comprehensively predict starch content in a complex environment with multiple factors.

[0074] S3. Construct a coupling model based on the main factor combination and constraint function to obtain the coupling influence of the main factor combination on the performance of the composite material.

[0075] It should be noted that constructing a coupling model based on the main factor combination and the constraint function to obtain the degree of coupling influence of the main factor combination on the performance of the composite material includes the following steps: defining a constraint function to describe the goal or requirement of the composite material performance; the input of the constraint function is a row of the factor combination matrix (i.e., a factor combination), and the output is the corresponding composite material performance index; using the multivariate function model to take the main factor combination as input and the output of the constraint function as the objective function to establish a coupling model; using the constructed coupling model to evaluate the factor combination to obtain the degree of coupling influence of the main factor combination on the composite material performance, and calculating different factor combinations through model solving to obtain the corresponding composite material performance index.

[0076] S4. An inversion model is established based on the degree of coupling influence, and the inversion model is used to perform inversion prediction on the starch content in the composite material.

[0077] Among them, establishing an inversion model based on the degree of coupling influence, and using the inversion model to inversely predict the starch content in the composite material includes the following steps: S41. Calculate the first-order derivative and the second-order derivative based on the degree of coupling influence, and construct the main factor parameters and the original near-infrared spectrum parameters.

[0078] S42. Perform correlation analysis between the original near-infrared spectral parameters and the main factor parameters, and use the cross-validation method to determine the sensitive parameters for inverting starch content.

[0079] It should be noted that the correlation analysis of the original near-infrared spectral parameters and the main factor parameters and the use of the cross-validation method to determine the sensitive parameters for inverting the starch content include the following steps: using the Pearson correlation coefficient to calculate the correlation between the original near-infrared spectral parameters and the main factor parameters, and by analyzing the size and direction of the correlation coefficient, determining which parameters have a strong correlation with the starch content; based on the results of the correlation analysis, selecting parameters that have a strong correlation with the starch content as sensitive parameters; in order to determine the performance of the selected parameters in inverting the starch content, using the cross-validation method for evaluation.

[0080] S43. Construct an inversion model based on sensitive parameters, and constrain the relative weight ratio in the inversion model according to prior information.

[0081] S431. Construct an inversion model based on the sensitive parameters of the inverted starch content, determine the constraint endpoint values ​​of the inversion model, and simultaneously determine the prior information of the main factor parameters and the original near-infrared spectrum parameters.

[0082] It should be noted that the inversion model is constructed according to the sensitive parameters of the inverted starch content, and the constraint endpoint values ​​of the inversion model are determined, and the prior information of the main factor parameters and the original near-infrared spectral parameters is determined, which includes the following steps: according to the correlation analysis and cross-validation results, sensitive parameters that have a strong correlation with the starch content and play a key role in the inversion model are selected; according to the selected sensitive parameters, the inversion model is constructed using support vector regression to predict the starch content; the selected sensitive parameters are used as input features and the starch content is used as the target variable for model training; in the inversion model, the prediction results of the model are limited by setting the constraint endpoint values, and the constraint endpoint values ​​can be determined according to actual needs and prior knowledge; incorporating prior information into the model training and prediction process can improve the stability and accuracy of the model.

[0083] S432. Add constraints to the relative weights of the inversion model based on prior information, and establish a constrained composite model.

[0084] S433, using a global optimization algorithm to determine the initial parameter values ​​of the inversion model, and setting a solution search method, and updating the relative weight value and parameter update value of the inversion model in the constrained composite model.

[0085] S434, taking the relative weight value and the parameter update value as the constraint function of the inversion model, when the difference between adjacent constraint functions is less than a preset threshold, the iterative process is stopped, otherwise, steps S433 to S434 are repeated.

[0086] S435, continuously updating the relative weight ratio and the parameter update value to obtain the final relative weight ratio and the final parameter update value.

[0087] It should be noted that after the present invention constructs the inversion model based on sensitive parameters, the relative weights in the inversion model are constrained according to prior information. The prior information can provide prior knowledge about the relative importance of parameters. By constraining the relative weights, it can ensure that the model is more stable and reliable in the prediction process, and reduce the impact of excessive or too small parameter weights on the results. This helps to reduce the overfitting or underfitting problems of the model and improve the generalization ability of the model. Furthermore, by constraining the relative weights, parameters that contribute less to the results can be eliminated, reducing the complexity and computational complexity of the model, and improving the efficiency and interpretability of the model.

[0088] S44. Based on the constraint condition of relative weight ratio, the inversion model is used to perform inversion prediction on the starch content in the composite material.

[0089] In summary, with the aid of the above-mentioned technical scheme of the present invention, the present invention can obtain the absorption, reflection or transmission information of the composite material sample at different wavelengths through spectral measurement technology, and provide rich multivariate information, which is helpful to fully understand the distribution state of the starch content in the composite material, and judge the main factors affecting the starch content from the distribution state, which is helpful to deeply understand the law of starch content distribution and the mechanism of action of the influencing factors, and then provide data basis for the subsequent inversion and identification of starch content in the composite material; the present invention can help to systematically generate various factor combinations through the matrix combination algorithm, and at the same time, the response surface method can efficiently screen out the main factors that have a significant impact on the target variable, which helps to improve the operating efficiency, and at the same time, combined with the model establishment of the response surface method, it can make a more detailed evaluation of the main factors and their interactions, so as to help to deeply understand the degree of influence and mode of action of each factor on the target variable, and then it can obtain sufficient information to determine The main factors are combined to reduce the operation cost and time; the present invention establishes an inversion model through the coupling influence degree between the main factors combination on the composite material performance, so as to comprehensively consider the complex influence of each main factor on the starch content, and more comprehensively explore the interaction effect between different factors. By considering these interactions, the accuracy of the inversion model is improved, thereby avoiding the inversion model from being interfered by secondary factors, and improving the calculation efficiency and prediction accuracy of the inversion model; the inversion model provided by the present invention can handle the inversion of starch content in complex composite materials, has strong applicability, and can provide a global perspective on the overall starch content distribution, thereby helping to discover the change trend of different regions or stages of the composite material. At the same time, the inversion model can accurately predict and estimate the starch content in the composite material, which helps to adjust the ratio of raw materials and processing parameters, so as to accurately obtain the required starch content, thereby improving production efficiency and production quality.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing an inverse model of starch content in composite materials based on response surface methodology, characterized in that: The method for constructing an inverse model of starch content in composite materials based on response surface methodology comprises the following steps: S1. Analyze the distribution state and distribution law of starch content in the composite material by using spectral measurement technology, and determine the main factors affecting the starch content in the composite material; S2, combining the response surface methodology with the matrix combination algorithm to determine the main factor combination; S3. Construct a coupling model based on the main factor combination and the constraint function to obtain the coupling influence of the main factor combination on the performance of the composite material; S4. An inversion model is established based on the degree of coupling influence, and the inversion model is used to perform inversion prediction on the starch content in the composite material.

2. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 1, characterized in that: The method of analyzing the distribution state and distribution law of starch content in the composite material by using spectral measurement technology and determining the main factors affecting the starch content in the composite material comprises the following steps: S11, collecting original near infrared spectrum data of the composite material using a near infrared spectrometer; S12, using a spectral feature selection algorithm to extract effective features from the original near-infrared spectral data, and dividing the composite materials into different clustering methods according to the classification index; S13, respectively screening spectral feature intervals for different clustering methods, and performing set operations on the spectral feature intervals to generate a joint matrix; S14, using an orthogonal linear discriminant analysis algorithm to transform the original near-infrared spectral data after feature extraction into an orthogonal linear discriminant analysis space, and generate a projection matrix; S15, performing similarity measurement calculation on the projection matrix, and generating an optimal spectral feature subset based on the similarity measurement calculation result; S16. Analyze the distribution state and distribution law of starch content in the composite material based on the optimal feature subset, and extract spectral feature representation therefrom; S17. Analyze the main factors affecting starch content in the composite material based on the spectral characteristics.

3. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 2, characterized in that: The method of transforming the original near-infrared spectral data after feature extraction into the orthogonal linear discriminant analysis space by using the orthogonal linear discriminant analysis algorithm and generating a projection matrix includes the following steps: S141, adding historical near infrared spectrum data of the same type of composite materials to the original near infrared spectrum data after feature extraction to obtain fused near infrared spectrum data; S142, performing partial least square feature extraction on the fused near-infrared spectrum data to obtain a least square feature matrix; S143, performing orthogonal linear discriminant feature extraction on the least square feature matrix in the orthogonal linear discriminant analysis space to obtain an orthogonal linear discriminant feature matrix; S144, converting the orthogonal linear discriminant feature matrix into a solution optimization matrix, sorting the solution optimization matrix, and forming a projection matrix with the feature vectors at the front of the sorting column.

4. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 3, characterized in that: The method of performing similarity measurement calculation on the projection matrix and generating an optimal spectral feature subset based on the similarity measurement calculation result comprises the following steps: S151, extracting feature points from the fused near-infrared spectrum data, calculating the distances between the feature points, and connecting each feature point with its nearest adjacent feature point to generate an adjacent undirected graph; S152, assigning a weight to each feature point in the adjacent undirected graph according to the distance between the feature points, and transforming the objective function by combining the weight with the Laplace matrix; S153, solving the target value based on the converted target function, obtaining the target vector corresponding to each feature point, and combining the target vectors to form a final projection matrix; S154, solving the final projection matrix to obtain a low-dimensional embedding matrix, and using the Euclidean distance to measure the similarity between the feature points in the low-dimensional embedding matrix space, and then constructing the optimal spectral feature subset based on the similarity between the feature points.

5. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 4, characterized in that: Determining the main factor combination by combining the response surface method with the matrix combination algorithm comprises the following steps: S21, respectively formulate the fitting relationship between starch content and each main factor, and generate combination categories using a matrix combination algorithm, and then calculate the dynamic change data of starch content in each combination category; S22, constructing a response surface model based on the dynamic change data, and using the response surface model to construct a polynomial response function between starch content and main factors; S23, using a data optimization iterative algorithm and Latin hypercube sampling to search for a minimum value in the polynomial response function, and using the minimum value as a parameter value; S24, generating a matrix classification combination between starch content and each main factor, and extracting the main factor combination from the matrix classification combination using a global optimization algorithm.

6. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 5, characterized in that: The steps of respectively formulating the fitting relationship between the starch content and each main factor, generating a combination category by using a matrix combination algorithm, and then calculating the dynamic change data of the starch content in each combination category include the following steps: S211, calculating the fitting relationship and preference degree between starch content and each main factor, and generating a combination category of starch content and each main factor; S212, analyzing the semantic relationship in the combined category space and the connection between the hierarchical structures, and calculating the semantic distance in the combined category space; S213, constructing a distance feature matrix model and a combined feature matrix model based on a combined category space based on semantic distance; S214, integrating the generated feature matrix model and the combined feature matrix model into the matrix decomposition model to obtain a matrix combination model based on the combined category space; S215. Calculate the dynamic change data of starch content in each combination category using a matrix combination model.

7. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 6, characterized in that: The analyzing of the semantic relationship in the combined category space and the connection between the hierarchical structures and calculating the semantic distance in the combined category space comprises the following steps: S2121. Calculate the jump span between each combination category and the hierarchical structure in the Hasse diagram of the combination category space; S2122: Determine the preference distance between the combination categories according to the jump span, and calculate the semantic distance between two adjacent combination categories that jump forward and backward according to the preference distance.

8. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 7, characterized in that: The expression of the distance feature matrix model is: ； In the formula, Indicates u The distance feature matrix of the main factors; represents the latent feature matrix of the u-th main factor; G represents the combined category space vector; T Indicates the total number of jumps; t Indicates t Jump times; A represents the indicator feature matrix; W The semantic distance between two adjacent combination categories representing the forward and backward jumps.

9. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 8, characterized in that: The method of establishing an inversion model based on the coupling influence degree and using the inversion model to perform inversion prediction on the starch content in the composite material comprises the following steps: S41. Calculate the first-order derivative and the second-order derivative based on the coupling influence degree, and construct the main factor parameters and the original near-infrared spectrum parameters; S42, performing correlation analysis on the original near infrared spectrum parameters and the main factor parameters, and determining the sensitive parameters for inverting starch content by using a cross-validation method; S43, constructing an inversion model based on sensitive parameters, and constraining relative weights in the inversion model according to prior information; S44. Based on the constraint condition of relative weight ratio, the inversion model is used to perform inversion prediction on the starch content in the composite material.

10. The method for constructing an inverse model of starch content in composite materials based on response surface methodology according to claim 9, characterized in that: The method of constructing an inversion model based on sensitive parameters and constraining the relative weight ratio in the inversion model according to prior information comprises the following steps: S431, constructing an inversion model according to the sensitive parameters of the inverted starch content, and determining the constraint endpoint values ​​of the inversion model, and determining the prior information of the main factor parameters and the original near infrared spectrum parameters; S432, adding constraints to the relative weight ratio of the inversion model according to the prior information, and establishing a constrained composite model; S433, using a global optimization algorithm to determine the initial parameter values ​​of the inversion model, and setting a solution search method, and updating the relative weight value and parameter update value of the inversion model in the constrained composite model; S434, taking the relative weight value and the parameter update value as the constraint function of the inversion model, when the difference between adjacent constraint functions is less than a preset threshold, the iterative process is stopped, otherwise, steps S433 to S434 are repeated; S435, continuously updating the relative weight ratio and the parameter update value to obtain the final relative weight ratio and the final parameter update value.

Images