Method and device for predicting flame retardancy of wood-based composite materials

Abstract

The present invention discloses a method and device for predicting the flame retardancy of wood-based composites, belonging to the field of forestry engineering. The present invention trains a machine learning model capable of predicting the flame retardancy of wood-based composites and finds out the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data. The present invention can be applied to the production practice of flame-retardant wood-based composites. According to the existing conditions such as the structural data and production process data of wood veneers, the actual value of their flame retardancy can be quickly predicted through the trained machine learning model, which has the advantages of being fast, efficient, and having high prediction accuracy, realizing rapid and accurate prediction of the performance of the final product. At the same time, through the aforementioned structure-activity relationship, the key factors affecting the flame retardancy can be clarified, guiding the optimization of raw materials and production processes, reducing the trial-and-error cost of enterprise processing, improving production efficiency, and further enhancing the processing level of flame-retardant wood-based composites.

Description

Technical Field

[0001] The present invention relates to the field of forestry engineering, and particularly to a method and device for predicting the flame retardancy of wood-based composites. Background Art

[0002] Wood-based composites are commonly used consumables in the fields of home furnishing, decoration, etc., and can be seen everywhere in daily life. When a fire occurs, the flame retardancy of wood materials can reduce the extent of fire spread, ensure life safety, reduce property losses, and even prevent problems before they occur.

[0003] There are many factors affecting the flame retardancy of wood-based composites, such as wood species, panel material and thickness, glue type, and glue application amount. In the actual production process, the processing procedures of flame-retardant wood-based composites are complex and more affected by experience. Workers often need rich experience to roughly judge which are the main factors affecting the flame retardancy of wood-based composites, and repeated experiments are required to achieve the desired performance effect. Summary of the Invention

[0004] To solve the defects of the prior art, the present invention provides a method and device for predicting the flame retardancy of wood-based composites, which realizes the rapid and accurate prediction of the flame retardancy of wood-based composites, guides the optimization of raw materials and production processes, reduces the processing trial-and-error cost of enterprises, and improves production efficiency.

[0005] The present invention provides the following technical solutions:

[0006] A method for predicting the flame retardancy of wood-based composites, the method comprising:

[0007] Establishing a reference data set including multiple wood veneer data samples; wherein each wood veneer data sample includes several features of the wood veneer, the types of the features include structural data and production process data, the wood veneer data sample is labeled with label data, and the label data includes flame retardancy data;

[0008] Digitizing and normalizing the reference data set;

[0009] Dividing the reference data set into a training set and a test set, training a machine learning model based on the random forest algorithm constructed by using the training set, and validating by using the test set to obtain a machine learning model capable of predicting the flame retardancy of wood-based composites;

[0010] Performing dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data.

[0011] Further, the establishment of a reference dataset including multiple wood veneer data samples includes:

[0012] Obtain the structural data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer; wherein, the structural data includes one or more of veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finish material, and finish material thickness;

[0013] Obtain the production process data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer; wherein, the production process data includes one or more of glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount;

[0014] Obtain the flame retardant performance data of the wood veneer after production, wherein the flame retardant performance data includes one or more of peak heat release rate, total heat release, oxygen index, and surface bonding strength;

[0015] Take the structural data and production process data as the features of the wood veneer data sample, and take the flame retardant performance data as the label data of the wood veneer data sample to obtain a reference dataset including multiple wood veneer data samples.

[0016] Further, the dimensionality reduction fitting of the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardant performance of the wood-based composite material and its structural data and production process data includes:

[0017] Calculate the feature mean of each feature of the reference dataset

[0018]

[0019] Wherein, is the feature mean of the j-th feature of the reference dataset, is the j-th feature of the i-th wood veneer data sample of the reference dataset, i = 1, 2, …, m, m is the total number of wood veneer data samples, j = 1, 2, …, n, n is the total number of feature types;

[0020] Use the feature mean of each feature of the reference dataset To update the reference dataset;

[0021]

[0022] Where D * is the updated reference dataset, and y j represents the updated reference dataset D* The j-th feature of

[0023] Based on the updated reference dataset D * Establish the covariance matrix B;

[0024]

[0025] Wherein, y a and y b respectively represent the a-th and b-th features of the updated reference dataset D * ; and respectively represent the a-th and b-th features of the i-th wood veneer data sample in the updated reference dataset D * ; and respectively represent the mean values of the a-th and b-th features of the updated reference dataset D * ;

[0026] Determine the dimension p of the features after dimensionality reduction, perform eigenvalue decomposition on the covariance matrix B to obtain eigenvalues λ = {λ1, λ2,..., λ n} and eigenvectors μ = {μ1, μ2,..., μ n};

[0027] Calculate the principal component P corresponding to each eigenvector j ;

[0028] P j =(μ j ) T B

[0029] Solve the contribution rate W of the principal components corresponding to the n eigenvalues j ;

[0030]

[0031] Combine all the contribution rates W of the principal components j to form a set W = {W1, W2,..., W n}, and obtain the structure-activity relationship between the flame retardancy of the wood-based composite material and its structural data and production process data.

[0032] Furthermore, the step of dividing the reference dataset into a training set and a test set, and training the constructed machine learning model based on the random forest algorithm using the training set includes:

[0033] Divide the reference dataset into a training set and a test set according to a ratio of 6:4;

[0034] Build a machine learning model based on the random forest algorithm, where the machine learning model includes a set number of decision trees, and the tree depth of each decision tree is a set depth;

[0035] Use the training set to train and tune the parameters of the machine learning model;

[0036] Adopt the K-fold cross-validation method to re-divide the training set and the test set, and repeatedly train and tune the parameters of the machine learning model until a machine learning model that can predict the flame retardancy of wood-based composites is obtained.

[0037] Further, the building of the machine learning model based on the random forest algorithm includes:

[0038] Randomly extract a set number of features from the reference dataset;

[0039] Calculate the Gini coefficient of each selected feature, and determine the feature that serves as the root node of the decision tree according to the Gini coefficient;

[0040] Randomly select from the remaining selected features to construct the remaining part of the decision tree; and iteratively construct each decision tree until all decision trees are constructed;

[0041] Perform iterative loops to construct multiple groups of decision trees by randomly extracting features multiple times, and determine the final decision tree through voting statistics.

[0042] Further, the digitalization and normalization of the reference dataset include:

[0043] Digitize the data in the reference dataset through Python technology;

[0044] Normalize the format and unit of the data in the reference dataset through Python technology.

[0045] Further, before dividing the reference dataset into a training set and a test set and using the training set to train the constructed machine learning model based on the random forest algorithm, it also includes:

[0046] Use the evidential reasoning rule to conduct a quality assessment on the digitized and normalized reference dataset, and evaluate the integrity and consistency of the reference dataset;

[0047] Conduct data cleaning on the reference dataset after quality assessment to remove outliers; conduct quality assessment on the reference dataset after data cleaning using the evidential reasoning rule again until its integrity and consistency meet the requirements.

[0048] Further, the quality of the digitized and normalized reference data set is evaluated using the evidential reasoning rule to evaluate the integrity and consistency of the reference data set, including:

[0049] Divide the reference data set into K sample subsets of equal size;

[0050] Traverse each of the sample subsets in turn, and use the current sample subset of each traversal as the validation set, and the remaining wood veneer data samples as the training set for quality evaluation;

[0051] Use the statistical values of the evaluation metrics obtained from the K - time quality evaluations as the final evaluation metrics, and evaluate the integrity and consistency of the reference data set through the final evaluation metrics.

[0052] Further, the reference data set after quality evaluation is subjected to data cleaning to remove outliers, including:

[0053] Represent the reference data set as a two - dimensional array, and for the element at any position in the reference data set, generate an observation window of a set size centered on the element at that position;

[0054] Calculate the median and standard deviation of all elements within the observation window;

[0055] If the absolute value of the difference between the element at any position and the median exceeds a set multiple of the standard deviation, then use the median to replace the element at that position.

[0056] Further, the establishment of a reference data set including multiple wood veneer data samples includes:

[0057] A device for predicting the flame - retardant performance of a wood - based composite material, the device includes:

[0058] A reference data set establishment module for establishing a reference data set including multiple wood veneer data samples; wherein, each wood veneer data sample includes several features of the wood veneer, the types of the features include structural data and production process data, and the wood veneer data sample is labeled with label data, and the label data includes flame - retardant performance data;

[0059] A pre - processing module for digitizing and normalizing the reference data set;

[0060] A model training module for dividing the reference data set into a training set and a test set, training a machine learning model based on the random forest algorithm constructed using the training set, and validating using the test set to obtain a machine learning model capable of predicting the flame - retardant performance of the wood - based composite material;

[0061] The structure-activity relationship determination module is used to perform dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of the wood composite material and its structural data and production process data.

[0062] Further, the reference dataset establishment module includes:

[0063] The structural data acquisition unit is used to acquire the structural data of the wood veneer collected with the wood veneer as the basic unit during the production process of the wood veneer; wherein, the structural data includes one or more of veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finish material, and finish material thickness.

[0064] The production process data acquisition unit is used to acquire the production process data of the wood veneer collected with the wood veneer as the basic unit during the production process of the wood veneer; wherein, the production process data includes one or more of glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount.

[0065] The flame retardancy performance data acquisition unit is used to acquire the flame retardancy performance data of the produced wood veneer, wherein the flame retardancy performance data includes one or more of peak heat release rate, total heat release, oxygen index, and surface bonding strength.

[0066] The reference dataset establishment unit is used to use the structural data and production process data as the features of the wood veneer data samples, and the flame retardancy performance data as the label data of the wood veneer data samples to obtain a reference dataset including multiple wood veneer data samples.

[0067] Further, the structure-activity relationship determination module includes:

[0068] The mean calculation unit is used to calculate the feature mean of each feature of the reference dataset

[0069]

[0070] wherein, is the feature mean of the j-th feature of the reference dataset, is the j-th feature of the i-th wood veneer data sample of the reference dataset, i = 1, 2, …, m, m is the total number of wood veneer data samples, j = 1, 2, …, n, n is the total number of feature types;

[0071] The update unit is used to use the feature mean of each feature of the reference dataset to update the reference dataset;

[0072]

[0073] Among them, D * is the updated reference data set, and y j represents the j-th feature of the updated reference data set D * .

[0074] Covariance matrix establishment unit, used to establish a covariance matrix B according to the updated reference data set D * .

[0075]

[0076] Among them, y a and y b respectively represent the a-th and b-th features of the updated reference data set D * , and respectively represent the a-th and b-th features of the i-th wood veneer data sample of the updated reference data set D * , and respectively represent the mean values of the a-th and b-th features of the updated reference data set D * .

[0077] Eigenvalue decomposition unit, used to determine the dimension p of the features after dimensionality reduction, perform eigenvalue decomposition on the covariance matrix B, and obtain eigenvalues λ = {λ1, λ2,..., λ n} and eigenvectors μ = {μ1, μ2,..., μ n};

[0078] Principal component calculation unit, used to calculate the principal component P corresponding to each eigenvector j ;

[0079] P j =(μ j ) T B

[0080] Contribution rate calculation unit, used to solve the principal component contribution rate W corresponding to n eigenvalues j ;

[0081]

[0082] Structure-activity relationship determination unit, used to form a set W = {W1, W2,..., W j} with all the principal component contribution rates W n , and obtain the structure-activity relationship between the flame retardant performance of wood-based composites and their structural data and production process data.

[0083] Further, the model training module includes:

[0084] A data partitioning unit for partitioning the reference data set into a training set and a test set according to a ratio of 6:4;

[0085] A random forest building unit for building a machine learning model based on the random forest algorithm, where the machine learning model includes a set number of decision trees, and the tree depth of each decision tree is a set depth;

[0086] A model training unit for training and tuning the parameters of the machine learning model using the training set;

[0087] An iterative training unit for re-partitioning the training set and the test set using the K-fold cross-validation method, and repeatedly training and tuning the parameters of the machine learning model until a machine learning model capable of predicting the flame retardant performance of wood-based composites is obtained.

[0088] Further, the random forest building unit includes:

[0089] A data extraction sub-unit for randomly extracting a set number of features from the reference data set;

[0090] A root node building sub-unit for calculating the Gini coefficient of each selected feature, and determining the feature serving as the root node of the decision tree according to the Gini coefficient;

[0091] A decision tree building sub-unit for randomly selecting from the remaining selected features to construct the remaining part of the decision tree; and cyclically iterating to construct each decision tree until all decision trees are constructed;

[0092] A cyclic iteration sub-unit for performing cyclic iteration, constructing multiple groups of decision trees by randomly extracting features multiple times, and determining the final decision tree by voting statistics.

[0093] Further, the preprocessing module includes:

[0094] A digitization unit for digitizing the data in the reference data set through Python technology;

[0095] A normalization unit for normalizing the format and unit of the data in the reference data set through Python technology.

[0096] Further, the device further includes:

[0097] A data evaluation module for evaluating the quality of the digitized and normalized reference data set using the evidential reasoning rule, and evaluating the integrity and consistency of the reference data set;

[0098] A data cleaning module is used to clean the reference data set after quality assessment to remove outliers; and to use the evidential reasoning rule to perform quality assessment on the reference data set after data cleaning until its integrity and consistency meet the requirements.

[0099] Furthermore, the data evaluation module includes:

[0100] A sample subset division unit is used to divide the reference data set into K sample subsets of equal size;

[0101] A traversal training unit is used to sequentially traverse each of the sample subsets, and use the current sample subset of each traversal as the validation set, and the remaining wood veneer data samples as the training set to perform quality assessment;

[0102] An evaluation unit is used to use the statistical values of the evaluation metrics obtained from the K times of quality assessment as the final evaluation metric, and evaluate the integrity and consistency of the reference data set through the final evaluation metric.

[0103] Furthermore, the data cleaning module includes:

[0104] An observation window generation unit is used to represent the reference data set in a two-dimensional array, and for an element at any position in the reference data set, generate an observation window of a set size centered on the element at that position;

[0105] An observation window calculation unit is used to calculate the median and standard deviation of all elements within the observation window;

[0106] A judgment unit is used to, if the absolute value of the difference between the element at the any position and the median exceeds a set multiple of the standard deviation, replace the element at the any position with the median.

[0107] The present invention has the following beneficial effects:

[0108] The present invention trains a machine learning model capable of predicting the flame retardancy of wood-based composites and identifies the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data. The present invention can be applied to the production practice of flame-retardant wood-based composites. According to the existing conditions such as the structural data and production process data of wood veneers, the actual value of their flame retardancy can be quickly predicted through the trained machine learning model, which has the advantages of being fast, efficient, and having high prediction accuracy, realizing the rapid and accurate prediction of the performance of the final product. At the same time, through the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data, the key factors affecting the flame retardancy can be identified, which can be used to guide the optimization of raw materials and production processes, reduce the trial-and-error cost of enterprise processing, improve production efficiency, and further enhance the processing level of flame-retardant wood-based composites. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Figure 1 is a flowchart of the method for predicting the flame retardancy of wood-based composites of the present invention;

[0110] Figure 2 is a schematic diagram of a random forest;

[0111] Figure 3 is a schematic diagram of the device for predicting the flame retardancy of wood-based composites of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0112] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the present invention to be protected, but only represents the selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0113] The embodiments of the present invention provide a method for predicting the flame retardancy of wood-based composites, as Figure 1 shown, the method includes:

[0114] S100: Establish a reference data set including multiple wood veneer data samples; wherein, each wood veneer data sample includes several features of the wood veneer, and the types of the features include structural data and production process data, and the wood veneer data sample is labeled with label data, and the label data includes flame retardancy data.

[0115] The structural data and production process data of wood veneers are factors affecting the flame retardancy of wood-based composites. The structural data includes veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finishing material, and finishing material thickness, etc. The production process data includes glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount, etc. The flame retardancy data includes peak heat release rate, total heat release, oxygen index, and surface bonding strength, etc. The present invention obtains 12 characteristics affecting the flame retardancy of wood-based composites and obtains 4 types of data representing flame retardancy, and infers 4 types of flame retardancy data through 12 characteristics.

[0116] In the actual production process of wood veneers, based on the wood veneer type, data collection is carried out on the above 16 types of data. The above 14 characteristics of each wood veneer are used as the characteristics of a data sample, and the 4 types of flame retardancy data are used as label data. A total of about 600 data samples are collected to form a reference data set.

[0117] S200: Digitalize and normalize the reference data set.

[0118] The data in the collected wood veneer data samples includes Chinese characters, English characters, and numerical characters, etc. For data such as Chinese characters and English characters, it is necessary to digitalize and normalize them into digital data recognizable by a computer.

[0119] S500: Divide the reference data set into a training set and a test set, use the training set to train the machine learning model based on the random forest algorithm, and use the test set for verification to obtain a machine learning model capable of predicting the flame retardancy of wood-based composites.

[0120] The machine learning model used in the present invention is a regression model based on the random forest algorithm. By using the training set to train the machine learning algorithm regression model and using the test set to test the results of the algorithm, a machine learning model capable of predicting the flame retardancy of wood-based composites can be obtained.

[0121] S600: Perform dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data.

[0122] The present invention trains a machine learning model capable of predicting the flame retardant performance of wood-based composites and finds out the structure-activity relationship between the flame retardant performance of wood-based composites and their structural data and production process data. The present invention can be applied to the production practice of flame retardant wood-based composites. According to the existing conditions such as the structural data and production process data of wood veneers, the actual value of the flame retardant performance can be quickly predicted through the trained machine learning model, which has the advantages of being fast, efficient, and having high prediction accuracy, realizing the rapid and accurate prediction of the performance of the final product. At the same time, through the structure-activity relationship between the flame retardant performance of wood-based composites and their structural data and production process data, the key factors affecting the flame retardant performance can be clarified, which can be used to guide the optimization of raw materials and production processes, reduce the trial-and-error cost of enterprise processing, improve production efficiency, and further improve the processing level of flame retardant wood-based composites.

[0123] As an improvement of an embodiment of the present invention, the foregoing S100 includes:

[0124] S101: Obtain the structural data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer.

[0125] Among them, the structural data includes one or more of the foregoing veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, decorative material, and decorative material thickness.

[0126] S102: Obtain the production process data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer.

[0127] Among them, the production process data includes one or more of the foregoing glue types, glue application amount, hot pressing temperature, flame retardant types, and flame retardant addition amount.

[0128] S103: Obtain the flame retardant performance data of the produced wood veneer.

[0129] Among them, the flame retardant performance data includes one or more of the foregoing peak heat release rate, total heat release, oxygen index, and surface bond strength;

[0130] S104: Take the structural data and production process data as the features of the wood veneer data sample, and take the flame retardant performance data as the label data of the wood veneer data sample to obtain a reference data set including multiple wood veneer data samples.

[0131] The present invention does not limit the method of digitizing and normalizing the reference data set. One example includes:

[0132] S201: Digitize the data in the reference data set through Python technology.

[0133] S202: Normalize the format and units of the data in the reference dataset through Python technology.

[0134] During digitization and normalization, Chinese and English information such as tree species names and material names can be digitized through Python and stored in CSV format. For example, for data with Chinese characters and English characters such as tree species names and flame retardant types, the decimal value of the ASCII code of the first letter of its Chinese pinyin or English is used as the result value. For example, for poplar wood, the ASCII code value of its first letter is 89.

[0135] During the process of the factory collecting data, there is a lot of useless data, and most of it is qualified data obtained based on production experience. Therefore, the data needs to be further processed. The method of the present invention further includes:

[0136] S300: Use the evidence reasoning rule to perform quality assessment on the digitized and normalized reference dataset, and evaluate the integrity and consistency of the reference dataset.

[0137] This step is used to perform quality assessment on the reference dataset and provide a basis for data availability, specifically as follows:

[0138] S301: Divide the reference dataset into K sample subsets of equal size.

[0139] S302: Traverse each sample subset in turn, and use the current sample subset of each traversal as the validation set, and the remaining wood veneer data samples as the training set to establish a model and perform quality assessment;

[0140] S303: Use the statistical values of the evaluation indicators obtained from the K - time quality assessment as the final evaluation indicators, and evaluate the integrity and consistency of the reference dataset through the final evaluation indicators.

[0141] The present invention uses the K - fold cross - validation method to repeatedly divide the reference dataset, perform multiple evaluations, and use statistical values such as the variance or average value of the multiple evaluation results as the final evaluation indicators to verify the integrity and consistency of the reference dataset.

[0142] S400: Perform data cleaning on the reference dataset after quality assessment to remove outliers; use the evidence reasoning rule to perform quality assessment on the reference dataset after data cleaning again until its integrity and consistency meet the requirements.

[0143] Data cleaning is used to remove noise during the collection process and reduce the impact of noise on the computer prediction model. The specific implementation methods include:

[0144] S401: Represent the reference dataset as a two - dimensional array, and for any element at any position in the reference dataset, generate an observation window of a set size centered on this element.

[0145] Exemplarily, the reference data set is represented by a two-dimensional array D = {x 1 , x 2 ,... x m} T where x i represents the i-th wood veneer data sample of the reference data set D, represents the j-th feature of the i-th wood veneer data sample of the reference data set D, i = 1, 2,..., m, where m is the total number of wood veneer data samples, and j = 1, 2,..., n, where n is the total number of feature types. The two-dimensional array D = {x 1 , x 2 ,... x m} T is expanded, and its form is as follows:

[0146]

[0147] For any element at a position a observation window of a set size is generated centered on it.

[0148] S402: Calculate the median and standard deviation of all elements within the observation window.

[0149] Assume that the size of the entire observation window is 2r + 1. Arrange all elements in the observation window in ascending order, calculate the median, that is, the (r + 1)-th number after sorting, and calculate the standard deviation.

[0150] S403: If the absolute value of the difference between any element and the median exceeds a set multiple of the standard deviation, then use the median to replace the element at any position.

[0151] Assume that the set multiple is 3 times. If the absolute value of the difference between the element and the median exceeds 3 times the standard deviation, then replace the element with this median.

[0152] As another improvement of the embodiment of the present invention, the foregoing S500 includes:

[0153] S501: Divide the reference data set after data cleaning into a training set and a test set according to a ratio of 6:4.

[0154] For the foregoing reference data set of about 600 wood veneer data samples, after division, about 360 sample data are used as the training set, and 240 sample data are used as the test set.

[0155] S502: Establish a machine learning model based on the random forest algorithm, where the machine learning model includes a set number of decision trees, and the tree depth of each decision tree is a set depth.

[0156] Exemplarily, as Figure 2 shown, the number of decision trees is 100, and the tree depth of each decision tree is 3. The construction process of the random forest is as follows:

[0157] 1. Randomly extract a set number of features from the reference dataset after data cleaning.

[0158] 2. Calculate the Gini coefficient of each selected feature, and determine the feature as the root node of the decision tree according to the Gini coefficient. The Gini coefficient can be calculated by the following formula.

[0159]

[0160] where G represents the Gini coefficient, and P j represents the probability of the j-th feature.

[0161] 3. Randomly select from the remaining selected features to construct the remaining part of the decision tree; and iteratively construct each decision tree until all decision trees are constructed.

[0162] 4. Iteratively construct multiple sets of decision trees by randomly extracting features multiple times, and determine the final decision tree by voting statistics.

[0163] S503: Train and tune the machine learning model using the training set.

[0164] S504: Re-partition the training set and the test set using the K-fold cross-validation method, and repeat training and tuning the machine learning model until a machine learning model that can predict the flame retardancy of wood-based composites is obtained.

[0165] The expected accuracy of the finally obtained machine learning model can exceed 95%. After the model is trained, the main factors affecting the flame retardancy can be fitted according to the principal component analysis algorithm (PCA, Principal components analysis), the regression function obtained by the model regression can be dimensionally reduced, and the structure-activity relationship formula between functions and materials can be formulated to analyze the structure-activity relationship of the flame retardancy of wood-based composites. The specific implementation process is as follows:

[0166] S601: Calculate the feature mean of each feature of the reference dataset

[0167]

[0168] where is the feature mean of the j-th feature of the reference dataset, The j-th feature of the i-th wood veneer data sample in the reference data set, where i = 1, 2, …, m (m is the total number of wood veneer data samples) and j = 1, 2, …, n (n is the total number of feature types).

[0169] S602: Use the feature mean of each feature in the reference data set to update the reference data set and centralize all features.

[0170]

[0171] where D * is the updated reference data set, and y j represents the j-th feature of the updated reference data set D * .

[0172] S603: Establish the covariance matrix B based on the updated reference data set D * .

[0173]

[0174] where y a and y b represent the a-th and b-th features of the updated reference data set D * respectively, and represent the a-th and b-th features of the i-th wood veneer data sample in the updated reference data set D * respectively, and and represent the mean values of the a-th and b-th features of the updated reference data set D * respectively. The diagonal of the covariance matrix B is the variance of the features, and the remaining non-diagonal elements are the covariances of the corresponding features.

[0175] S604: Determine the reduced feature dimension p according to the application requirements, perform eigenvalue decomposition on the covariance matrix B to obtain the eigenvalues λ = {λ1, λ2,..., λ n} and eigenvectors μ = {μ1, μ2,..., μ n}

[0176] S605: Calculate the principal component P j corresponding to each eigenvector.

[0177] P j = (μ j ) T B

[0178] S606: Solve the principal component contribution rate W corresponding to the n eigenvaluesj 。

[0179]

[0180] S607: Combine all the principal component contribution rates W j to form a set W = {W1, W2,..., W n}} to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data.

[0181] The embodiment of the present invention also provides a device for predicting the flame retardancy of wood-based composites, as Figure 3 shown. The device includes:

[0182] A reference data set establishment module 1 for establishing a reference data set including multiple wood veneer data samples; wherein each wood veneer data sample includes several features of the wood veneer, and the types of the features include structural data and production process data, and the wood veneer data samples are labeled with label data, and the label data includes flame retardancy data.

[0183] A preprocessing module 2 for digitizing and normalizing the reference data set.

[0184] A model training module 3 for dividing the reference data set into a training set and a test set, training a machine learning model based on the random forest algorithm constructed by using the training set, and validating by using the test set to obtain a machine learning model capable of predicting the flame retardancy of wood-based composites.

[0185] A structure-activity relationship determination module 4 for performing dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data.

[0186] The present invention has trained a machine learning model capable of predicting the flame retardancy of wood-based composites and found the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data. The present invention can be applied to the production practice of flame-retardant wood-based composites. According to the existing conditions such as the structural data and production process data of wood veneers, the actual value of the flame retardancy can be quickly predicted through the trained machine learning model, which has the advantages of being fast, efficient, and having high prediction accuracy, and realizes the rapid and accurate prediction of the performance of the final product. At the same time, through the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data, the key factors affecting the flame retardancy can be clarified, which can be used to guide the optimization of raw materials and production processes, reduce the trial-and-error cost of enterprise processing, improve production efficiency, and further improve the processing level of flame-retardant wood-based composites.

[0187] As an improvement of the embodiment of the present invention, the foregoing reference dataset establishment module includes:

[0188] A structural data acquisition unit, configured to acquire the structural data of the wood veneer, taking the wood veneer as a unit, during the production process of the wood veneer.

[0189] Wherein, the structural data includes one or more of veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finish material, and finish material thickness.

[0190] A production process data acquisition unit, configured to acquire the production process data of the wood veneer, taking the wood veneer as a unit, during the production process of the wood veneer.

[0191] Wherein, the production process data includes one or more of glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount.

[0192] A flame retardant performance data acquisition unit, configured to acquire the flame retardant performance data of the produced wood veneer.

[0193] Wherein, the flame retardant performance data includes one or more of peak heat release rate, total heat release, oxygen index, and surface bonding strength.

[0194] A reference dataset establishment unit, configured to use the structural data and production process data as the features of the wood veneer data sample, and use the flame retardant performance data as the label data of the wood veneer data sample, to obtain a reference dataset including multiple wood veneer data samples.

[0195] An example of the preprocessing module of the present invention includes:

[0196] A digitization unit, configured to digitize the data in the reference dataset through Python technology.

[0197] A normalization unit, configured to normalize the format and unit of the data in the reference dataset through Python technology.

[0198] The device of the present invention may further include:

[0199] A data evaluation module, configured to perform quality evaluation on the digitized and normalized reference dataset using the evidence reasoning rule, and evaluate the integrity and consistency of the reference dataset.

[0200] A data cleaning module, configured to perform data cleaning on the reference dataset after quality evaluation to remove outliers; and perform quality evaluation on the reference dataset after data cleaning again using the evidence reasoning rule until its integrity and consistency meet the requirements.

[0201] Among them, the data evaluation module includes:

[0202] A sample subset division unit for dividing the reference data set into K sample subsets of equal size.

[0203] A traversal training unit for sequentially traversing each sample subset, using the current sample subset of each traversal as the validation set, and the remaining wood veneer data samples as the training set for quality evaluation.

[0204] An evaluation unit for using the statistical values of the evaluation metrics obtained from K times of quality evaluations as the final evaluation metrics, and evaluating the integrity and consistency of the reference data set through the final evaluation metrics.

[0205] The data cleaning module includes:

[0206] An observation window generation unit for representing the reference data set as a two-dimensional array, and generating an observation window of a set size centered on the element at any position in the reference data set.

[0207] An observation window calculation unit for calculating the median and standard deviation of all elements within the observation window.

[0208] A judgment unit for replacing the element at any position with the median if the absolute value of the difference between the element at any position and the median exceeds a set multiple of the standard deviation.

[0209] As another improvement of the embodiment of the present invention, the foregoing model training module includes:

[0210] A data division unit for dividing the reference data set into a training set and a test set according to a ratio of 6:4.

[0211] A random forest establishment unit for establishing a machine learning model based on the random forest algorithm, where the machine learning model includes a set number of decision trees, and the tree depth of each decision tree is a set depth.

[0212] A model training unit for training and tuning the machine learning model using the training set.

[0213] An iterative training unit for re-dividing the training set and the test set using the K-fold cross-validation method, and repeatedly training and tuning the machine learning model until a machine learning model capable of predicting the flame retardant performance of wood-based composites is obtained.

[0214] Among them, the foregoing random forest establishment unit includes:

[0215] A data extraction subunit for randomly extracting a set number of features from the reference data set.

[0216] The root node creates a sub-unit for calculating the Gini coefficient of each selected feature, and determines the feature serving as the root node of the decision tree according to the Gini coefficient.

[0217] The decision tree creation sub-unit is used to randomly select from the remaining selected features to construct the remaining part of the decision tree; and iteratively construct each decision tree until all decision trees are constructed.

[0218] The iterative sub-unit is used to perform iterative loop, construct multiple groups of decision trees by randomly extracting features multiple times, and determine the final decision tree by voting statistics.

[0219] Further, the aforementioned structure-activity relationship determination module includes:

[0220] The mean calculation unit is used to calculate the feature mean of each feature of the reference data set

[0221]

[0222] Wherein, is the feature mean of the j-th feature of the reference data set, is the j-th feature of the i-th wood veneer data sample of the reference data set, i = 1, 2, …, m, m is the total number of wood veneer data samples, j = 1, 2, …, n, n is the total number of feature types.

[0223] The update unit is used to update the reference data set by using the feature mean of each feature of the reference data set Update the reference data set.

[0224]

[0225] Wherein, D * is the updated reference data set, y j represents the j-th feature of the updated reference data set D * of.

[0226] The covariance matrix creation unit is used to create a covariance matrix B according to the updated reference data set D * Create a covariance matrix B.

[0227]

[0228] Wherein, y a and y b respectively represent the a-th and b-th features of the updated reference data set D * of, and respectively represent the updated reference data set D *The a-th and b-th features of the i-th wood veneer data sample and respectively represent the feature means of the a-th and b-th features of the updated reference data set D * of.

[0229] Eigenvalue decomposition unit, used to determine the reduced feature dimension p, perform eigenvalue decomposition on the covariance matrix B, and obtain eigenvalues λ = {λ1, λ2,..., λ n} and eigenvectors μ = {μ1, μ2,..., μ n}.

[0230] Principal component calculation unit, used to calculate the principal component P corresponding to each eigenvector j .

[0231] P j =(μ j ) T B

[0232] Contribution rate calculation unit, used to solve the principal component contribution rate W corresponding to n eigenvalues j .

[0233]

[0234] Structure-activity relationship determination unit, used to combine all the principal component contribution rates W j to form a set W = {W1, W2,..., W n}, and obtain the structure-activity relationship between the flame retardant performance of the wood composite material and its structural data and production process data.

[0235] For the device provided in the above embodiment, the implementation principle and the technical effects generated correspond one by one to the embodiments of the foregoing method. For the sake of brief description, for the parts not mentioned in the embodiment of the device, reference may be made to the corresponding contents in the embodiments of the foregoing method. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the modules and units described in the device can all refer to the corresponding processes in the embodiments of the foregoing method, and will not be repeated here.

[0236] The method for predicting the flame retardant performance of the wood composite material described in the above embodiment of the present invention can implement the business logic through a computer program and record it on a storage medium. The storage medium can be read and executed by a computer to achieve the effects of the solution described in the method embodiment of this specification. Therefore, the embodiment of the present invention also provides a computer-readable storage medium for predicting the flame retardant performance of the wood composite material, including a memory for storing processor-executable instructions, and the instructions, when executed by the processor, implement the steps of the method for predicting the flame retardant performance of the wood composite material including the foregoing embodiments.

[0237] The storage medium may include a physical device for storing information, which is usually a medium that digitizes the information and then stores it in an electrical, magnetic or optical manner. The storage medium may include: a device that stores information in an electrical energy manner, such as various memories, such as RAM, ROM, etc.; a device that stores information in a magnetic energy manner, such as a hard disk, a floppy disk, a magnetic tape, a magnetic core memory, a magnetic bubble memory, a USB flash drive; a device that stores information in an optical manner, such as a CD or a DVD. Of course, there are other readable storage media, such as quantum memory, graphene memory, etc.

[0238] The storage medium described above may also include other implementation methods according to the description of the method embodiment. The implementation principle and technical effects produced by this embodiment are the same as those of the aforementioned method embodiment. For details, please refer to the description of the relevant method embodiment, and no further description will be given here.

[0239] The embodiment of the present invention also provides a device for predicting the flame retardant properties of wood composite materials, which may be a separate computer, or may include an actual operating device using one or more of the methods or one or more of the embodiments of this specification. The device for predicting the flame retardant properties of wood composite materials may include at least one processor and a memory storing computer executable instructions, and when the processor executes the instructions, any one or more steps of the method for predicting the flame retardant properties of wood composite materials are implemented.

[0240] The device described above may also include other implementation methods according to the description of the method embodiment. The implementation principle and technical effects produced by this embodiment are the same as those of the aforementioned method embodiment. For details, please refer to the description of the relevant method embodiment, and no further description will be given here.

[0241] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting the flame retardancy of wood-based composites, characterized in that, The method includes: Establishing a reference data set including multiple wood veneer data samples; wherein, each of the wood veneer data samples includes several features of the wood veneer, the types of the features include structural data and production process data, the wood veneer data samples are labeled with label data, and the label data includes flame retardancy data; Digitizing and normalizing the reference data set; Dividing the reference data set into a training set and a test set, training a machine learning model based on the random forest algorithm constructed by using the training set, and validating by using the test set to obtain a machine learning model capable of predicting the flame retardancy of wood-based composites; Performing dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data; The establishing of the reference data set including multiple wood veneer data samples includes: Obtaining the structural data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer; wherein, the structural data includes one or more of veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finish material, and finish material thickness; Obtaining the production process data of the wood veneer collected with the wood veneer as the unit during the production process of the wood veneer; wherein, the production process data includes one or more of glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount; Obtaining the flame retardancy data of the produced wood veneer, wherein the flame retardancy data includes one or more of peak heat release rate, total heat release, oxygen index, and surface bond strength; Taking the structural data and production process data as the features of the wood veneer data samples, and taking the flame retardancy data as the label data of the wood veneer data samples to obtain a reference data set including multiple wood veneer data samples.

2. The method for predicting the flame retardancy of the wood composite material according to claim 1, characterized in that The performing of dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardancy of wood-based composites and their structural data and production process data includes: Calculate the feature mean of each feature in the reference data set Among them, is the mean value of the j-th feature of the reference data set, is the j-th feature of the i-th wood veneer data sample of the reference data set, where i = 1, 2, …, m, m is the total number of wood veneer data samples, and j = 1, 2, …, n, n is the total number of feature types; The feature mean of each feature of the reference data set is used Update the reference data set; Among them, D * is the updated reference data set, and y j represents the j-th feature of the updated reference data set D * ; According to the updated reference data set D * Establish the covariance matrix B; Among them, a = 1, 2, …, n, b = 1, 2, …, n, y a and y b respectively represent the a-th and b-th features of the updated reference dataset D * ; and respectively represent the a-th and b-th features of the i-th wood veneer data sample in the updated reference dataset D * ; and respectively represent the feature means of the a-th and b-th features of the updated reference dataset D * ; Determine the dimensionality p of the features after dimensionality reduction, perform eigenvalue decomposition on the covariance matrix B to obtain eigenvalues λ = {λ1, λ2,..., λ n} and eigenvectors μ = {μ1, μ2,..., μ n}; Calculate the principal component P corresponding to each eigenvector j ; Solve the principal component contribution rate W corresponding to n eigenvalues j ; All the contribution rates of the main components W j are grouped to form a set W = {W1, W2,..., W n}, and the structure-activity relationship between the flame retardancy of the wood-based composite material and its structural data and production process data is obtained.

3. The method for predicting the flame retardancy of the wood composite material according to claim 1, characterized in that The dividing of the reference data set into a training set and a test set and training the machine learning model based on the random forest algorithm constructed by using the training set includes: Dividing the reference data set into a training set and a test set according to a ratio of 6:4; Establishing a machine learning model based on the random forest algorithm, the machine learning model includes a set number of decision trees, and the tree depth of each decision tree is a set depth; Training and tuning the parameters of the machine learning model by using the training set; Re-dividing the training set and the test set by using the K-fold cross-validation method, and repeating the training and tuning of the machine learning model until a machine learning model capable of predicting the flame retardancy of wood-based composites is obtained.

4. The method for predicting the flame retardancy of the wood composite material according to claim 3, wherein, The establishing of the machine learning model based on the random forest algorithm includes: Randomly extracting a set number of features from the reference data set; Calculate the Gini coefficient for each selected feature, and determine the feature serving as the root node of the decision tree according to the Gini coefficient; Randomly select from the remaining selected features to construct the remaining part of the decision tree; and iteratively construct each decision tree until all decision trees are constructed; Perform iterative looping, construct multiple sets of decision trees by randomly extracting features multiple times, and determine the final decision tree by means of voting statistics.

5. The method for predicting the flame retardancy of the wood composite material according to claim 1, characterized in that, The digitizing and normalizing the reference data set includes: Digitize the data in the reference data set through Python technology; Normalize the format and unit of the data in the reference data set through Python technology.

6. The method for predicting the flame retardancy of the wood-based composite material according to any one of claims 1-5, characterized in that, Before dividing the reference data set into a training set and a test set and using the training set to train the machine learning model constructed based on the random forest algorithm, it also includes: Perform quality assessment on the digitized and normalized reference data set using the evidential reasoning rule to evaluate the integrity and consistency of the reference data set; Perform data cleaning on the reference data set after quality assessment to remove outliers; perform quality assessment on the reference data set after data cleaning again using the evidential reasoning rule until its integrity and consistency meet the requirements.

7. The method for predicting the flame retardancy of the wood composite material according to claim 6, wherein The performing quality assessment on the digitized and normalized reference data set using the evidential reasoning rule to evaluate the integrity and consistency of the reference data set includes: Divide the reference data set into K sample subsets of equal size; Traverse each of the sample subsets in sequence, and use the current sample subset of each traversal as the validation set and the remaining wood veneer data samples as the training set to perform quality assessment; Use the statistical values of the evaluation metrics obtained from the K times of quality assessment as the final evaluation metrics, and evaluate the integrity and consistency of the reference data set through the final evaluation metrics.

8. The method for predicting the flame retardancy of the wood-based composite material according to claim 6, characterized in that The performing data cleaning on the reference data set after quality assessment to remove outliers includes: Represent the reference data set as a two-dimensional array, and for the element at any position in the reference data set, generate an observation window of a set size centered on the element at that any position; Calculate the median and standard deviation of all elements within the observation window; If the absolute value of the difference between the element at the any position and the median exceeds a set multiple of the standard deviation, use the median to replace the element at the any position.

9. A device for predicting the flame retardancy of a wood composite material, characterized in that, The device includes: A reference data set establishment module for establishing a reference data set including multiple wood veneer data samples; wherein, each of the wood veneer data samples includes several features of the wood veneer, the types of the features include structural data and production process data, and the wood veneer data samples are labeled with label data, and the label data includes flame retardant performance data; A preprocessing module for digitizing and normalizing the reference data set; A model training module, which is used to divide the reference data set into a training set and a test set, train a machine learning model based on the random forest algorithm constructed by using the training set, and verify by using the test set to obtain a machine learning model capable of predicting the flame retardant performance of wood-based composites; A structure-activity relationship determination module, which is used to perform dimensionality reduction fitting on the regression function of the trained machine learning model through the principal component analysis algorithm to obtain the structure-activity relationship between the flame retardant performance of wood-based composites and their structural data and production process data; The reference data set establishment module includes: A structural data acquisition unit, which is used to acquire the structural data of the wood veneer collected with the wood veneer as a basic unit during the production process of the wood veneer; wherein, the structural data includes one or more of veneer tree species, veneer thickness, number of plywood glue layers, face material, face material thickness, finish material, and finish material thickness; A production process data acquisition unit, which is used to acquire the production process data of the wood veneer collected with the wood veneer as a basic unit during the production process of the wood veneer; wherein, the production process data includes one or more of glue type, glue application amount, hot pressing temperature, flame retardant type, and flame retardant addition amount; A flame retardant performance data acquisition unit, which is used to acquire the flame retardant performance data of the produced wood veneer, wherein the flame retardant performance data includes one or more of peak heat release rate, total heat release, oxygen index, and surface bonding strength; A reference data set establishment unit, which is used to use the structural data and production process data as the features of the wood veneer data sample, and use the flame retardant performance data as the label data of the wood veneer data sample to obtain a reference data set including multiple wood veneer data samples.

Images