Formula gene-based biodegradable mulching film performance prediction method
Through deep learning methods, the formula and performance database of biodegradable mulch film is constructed, and the encoding network and graph convolutional network are used to solve the problem of difficult prediction of biodegradable mulch film performance, and the rapid and accurate performance prediction is achieved, and R&D efficiency and management are improved. The effect of the tool.
Patent Information
- Application Number
- CN202411284133.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2025-01-03
AI Technical Summary
The physical and chemical properties of biodegradable mulch are difficult to accurately predict, resulting in limited research and development speed. The existing methods rely on empirical judgment and lack the ability to quickly simulate and predict.
Based on deep learning methods, a database of material formula and physical and chemical performance is constructed, and a coding network, feature extraction subnet and prediction subnet are used to learn the deep relationships in the formula through multi-layer perceptron, self-attention mechanism and graph convolution network. Predict the physical and chemical properties of mulch materials.
It achieves rapid and accurate prediction of the physical and chemical properties of biodegradable mulch at different formula ratios, improves the predictive performance, is simple and efficient, reduces costs, and provides a basic tool for evaluation and management of degradable mulch.
Smart Images
Figure CN119252366A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a prediction algorithm for the physical and chemical properties of a biodegradable mulch film formula, and in particular to a performance prediction method for a biodegradable mulch film formula based on deep learning. Background Art
[0002] For a long time, ground film has played a positive role in the efficient production of crops due to its advantages of heat preservation, moisture retention, insect prevention and weed control. However, the large-scale use of traditional non-degradable plastic ground film and the lack of recycling links have led to increasingly serious residual pollution problems. The promotion and application of biodegradable ground film (BDM) is an effective path for the prevention and control of ground film pollution and sustainable agricultural development in my country. However, the main raw materials for the production of biodegradable ground film are currently diverse, ranging from biodegradable polyester to modified additives. The differences in chemical structure and physical and chemical properties of different raw materials, the proportion of raw material formulas and their compatibility make the physical and chemical properties of biodegradable ground films vary greatly. At present, the physical and chemical properties of biodegradable ground films are still difficult to accurately predict. Researchers can only make rough judgments based on the physical and chemical properties of the raw materials themselves and the experience accumulated in the long-term research and development of ground film formulas, which greatly hinders the research and development rate of biodegradable ground films.
[0003] The formula of biodegradable mulch includes: the synthesis of one or more of PBAT, PBST, PBSA, PLA, PPC, PPCP, PGA, PBS, PHA, PCL and anti-hydrolysis agent, anti-ultraviolet aging agent, cross-linking agent, masterbatch and other additives. When the formula of biodegradable mulch changes, it is necessary to quickly simulate and predict the physical and chemical properties of the material. Summary of the invention
[0004] Purpose of the invention: In order to solve the above problems, this application is based on the established database of different material formulas and physical and chemical properties, uses deep learning methods to learn the deep relationship between data, explores the intrinsic connection between the data to be predicted and the molecular characteristics, and quickly obtains the physical and chemical properties of the material formula.
[0005] Technical solution: A method for predicting the performance of biodegradable mulch films based on formula genes, comprising the following steps:
[0006] Collect information on biodegradable mulch materials, obtain performance data of different formula materials through synthesis experiments based on the raw material information, and build a database;
[0007] The biodegradable mulch material information includes the molecular structure and proportion of biodegradable polyester and modification additives in different formulas and the physical and chemical properties of the synthetic materials; the performance data of the materials with different formulas include the performance of the mulch material;
[0008] The K-Fold cross-validation method is used to divide the data, and the biodegradable mulch raw material information is used as input, and the performance data of different formula materials are used as output to construct a prediction model; the prediction model includes an encoding network, a feature extraction subnetwork and a prediction subnetwork; the encoding network uses a multi-layer perceptron to encode the components in the formula, and then obtains the implicit features of the formula; the feature extraction subnetwork uses a self-attention mechanism to perform global correlation learning on the implicit features to obtain the deep features of the samples, and the prediction subnetwork learns the prediction functions of each performance from the deep features through a linear layer.
[0009] Furthermore, the molecular structure of biodegradable polyester includes chain structure and condensed structure, wherein the chain structure includes chemical composition, including main chain chemical composition and end group chemical composition; configuration, including optical isomerism; sequence structure and average composition; molecular size, including molecular weight and molecular weight distribution.
[0010] The condensed structure includes: crystalline structure, including crystallinity, grain size, and lamellar thickness; orientation structure and orientation degree.
[0011] Furthermore, the molecular structure of the modification additive includes chemical composition, configuration, structure, and molecular size.
[0012] Furthermore, the properties of the mulch materials include thermodynamic properties, including glass transition temperature, melting temperature, crystallization temperature; rheology, melt index, color value, density, acid value, terminal hydroxyl content, thickness, light transmittance and UV shielding properties, mechanical properties, including tensile strength, elongation at break, flexural strength, flexural modulus, impact strength; water vapor and oxygen permeability.
[0013] Furthermore, data preprocessing includes statistical analysis of the synthetic formulas of one or more of PBAT, PBST, PBSA, PLA, PPC, PPCP, PGA, PBS, PHA, PCL and anti-hydrolysis agents, anti-ultraviolet aging agents, cross-linking agents, masterbatches and other additives, projecting the performance of different formulas into a high-dimensional space using an autoencoder method, and then calculating the Euclidean distance between different formulas, clustering different formulas based on the distance matrix using a local outlier factor LOF method, and then screening out all formula samples whose distances exceed 50% of the longest distance, so that the distribution of formula samples covers all materials and the distribution of different formula data conforms to the average distribution.
[0014] Furthermore, the coding network encodes the molecular structure data in the biodegradable mulch film raw material information, and on the one hand, converts the SMILES code of the molecular structure into an extended connection fingerprint code through hash coding; on the other hand, the molecular structure graph is converted into a molecular structure feature value using graph theory, and further converted into a graph vector code using a graph convolutional network; the extended connection fingerprint code and the graph vector code are spliced to obtain a combined code of the material;
[0015] The molecular structure characteristic values include atomic vectors, bond vectors, and global vectors.
[0016] Furthermore, the encoding network shares weights.
[0017] Furthermore, the molecular structure graph is converted into molecular structure feature values by using graph theory, and further converted into graph vector encoding by using graph convolution network as follows:
[0018] Graph neural networks are used to process molecular structures. Each molecule is represented as a graph, where atoms are nodes and chemical bonds are edges. Node features are updated through graph convolution layers. The graph convolution formula is as follows:
[0019]
[0020] In the formula, X is the node feature, and A is the adjacency matrix. I is the identity matrix, is the degree matrix, W is the weight matrix, σ is the nonlinear activation function, and Y is the deep feature of X.
[0021] Furthermore, the extended connection fingerprint code and the graph vector code are concatenated to obtain the combined code of the material. The feature concatenation is expressed as:
[0022] X final =Concat(X Hash ,X GNN )
[0023] Where, X final is the final result of the combined encoding, X Hash is the result of hash encoding, X GNN The graph vector result obtained by the graph convolutional network.
[0024] A computing device comprises: one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include a method for executing a biodegradable mulch film performance prediction method based on formula genes.
[0025] A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to perform a method for predicting the performance of a biodegradable mulch film based on a recipe gene.
[0026] Beneficial effects: The model can quickly predict the physical and chemical properties of biodegradable mulch films at different formula ratios. The deep learning method is used to obtain the correlation information between different components, and the prediction performance of the model is greatly improved. This method is simple, efficient and low-cost, and is expected to provide a basic tool for the evaluation, management and prediction of biodegradable mulch films. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 is a flow chart of a method according to an embodiment of the present invention;
[0028] Figure 2 A schematic diagram of a prediction model according to an embodiment of the present invention;
[0029] Figure 3 A schematic diagram of a coding network according to an embodiment of the present invention; DETAILED DESCRIPTION
[0030] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.
[0031] like Figure 1 As shown, the present invention provides a technical solution:
[0032] A method for predicting the performance of biodegradable mulch films based on formula genes comprises the following steps:
[0033] Collect information on biodegradable mulch materials, obtain performance data of different formula materials through synthesis experiments based on the raw material information, and build a database;
[0034] The biodegradable mulch material information includes the molecular structure and proportion of biodegradable polyester and modification additives in different formulas and the physical and chemical properties of the synthetic materials; the performance data of the materials with different formulas include the performance of the mulch material;
[0035] Prediction models such as Figure 2 As shown, the K-Fold cross-validation method is used to divide the data into 8:2 ratios, and the biodegradable mulch raw material information is used as input, and the performance data of different formula materials are used as output to construct a prediction model; the prediction model includes an encoding network, a feature extraction subnetwork and a prediction subnetwork; the encoding network uses a multi-layer perceptron to encode each component in the formula, and then obtains the implicit features of the formula; the feature extraction subnetwork uses a self-attention mechanism to perform global correlation learning on the implicit features to obtain the deep features of the samples, and the prediction subnetwork learns the prediction functions of each performance from the deep features through a linear layer.
[0036] Furthermore, the molecular structure of biodegradable polyester includes chain structure and condensed structure, wherein the chain structure includes chemical composition, including main chain chemical composition and end group chemical composition; configuration, including optical isomerism; sequence structure and average composition; molecular size, including molecular weight and molecular weight distribution.
[0037] The condensed structure includes: crystalline structure, including crystallinity, grain size, and lamellar thickness; orientation structure and orientation degree.
[0038] Furthermore, the molecular structure of the modification additive includes chemical composition, configuration, structure, and molecular size.
[0039] Furthermore, the properties of the mulch materials include thermodynamic properties, including glass transition temperature, melting temperature, crystallization temperature; rheology, melt index, color value, density, acid value, terminal hydroxyl content, thickness, light transmittance and UV shielding properties, mechanical properties, including tensile strength, elongation at break, flexural strength, flexural modulus, impact strength; water vapor and oxygen permeability.
[0040] Furthermore, data preprocessing includes statistical analysis of the formulas of PBAT, PLA, PPC, PPCP, PGA, PBS materials and their modified additives, projecting the performance of different formulas into a high-dimensional space using the autoencoder method, and then calculating the Euclidean distance between different formulas. Different formulas are clustered based on the distance matrix using the local outlier factor LOF method, and then all formula samples with a distance exceeding 50% of the longest distance are screened out, so that the distribution of formula samples covers all materials and the distribution of different formula data conforms to the average distribution.
[0041] Furthermore, the coding network encodes the molecular structure data in the biodegradable mulch film raw material information, and on the one hand, converts the SMILES code of the molecular structure into an extended connection fingerprint code through hash coding; on the other hand, the molecular structure graph is converted into a molecular structure feature value using graph theory, and further converted into a graph vector code using a graph convolutional network; the extended connection fingerprint code and the graph vector code are spliced to obtain a combined code of the material;
[0042] Figure 3 The above process is shown, Figure 3 Here k is the smallest unit of scale in the molecular structure.
[0043] The molecular structure characteristic values include atomic vectors, bond vectors, and global vectors.
[0044] Furthermore, the molecular structure graph is converted into molecular structure feature values by using graph theory, and further converted into graph vector encoding by using graph convolution network as follows:
[0045] Graph neural networks are used to process molecular structures. Each molecule is represented as a graph, where atoms are nodes and chemical bonds are edges. Node features are updated through graph convolution layers. The graph convolution formula is as follows:
[0046]
[0047] In the formula, X is the node feature, and A is the adjacency matrix. I is the identity matrix, is the degree matrix, W is the weight matrix, σ is the nonlinear activation function, and Y is the deep feature of X.
[0048] Furthermore, the extended connection fingerprint code and the graph vector code are concatenated to obtain the combined code of the material. The feature concatenation is expressed as:
[0049] X final =Concat(X Hash ,X GNN )
[0050] Where, X final is the final result of the combined encoding, X Hash is the result of hash encoding, X GNN The graph vector result obtained by the graph convolutional network.
[0051] The hyperparameters of the prediction model are as follows: 8 samples are trained in each batch, and the total number of iterations is 300, that is, batchsize=8,epoch=300; the loss function indicator is the mean square error MSE; the optimizer is AdamW; the optimizer step size is set to 0.004, that is, learningrate=0.004.
Claims
1. A method for predicting the performance of biodegradable mulch films based on formula genes, characterized in that: The following steps are involved: Collect information on biodegradable mulch materials, obtain performance data of different formula materials through synthesis experiments based on the raw material information, and build a database; The biodegradable mulch material information includes the molecular structure and proportion of biodegradable polyester and modification additives in different formulas and the physical and chemical properties of the synthetic materials; the performance data of the materials with different formulas include the performance of the mulch material; The K-Fold cross-validation method is used to divide the data, and the information of the raw materials of the biodegradable mulch film is used as input, and the performance data of different formula materials are used as output to construct a prediction model; the prediction model includes an encoding network, a feature extraction subnetwork and a prediction subnetwork; the encoding network uses a multi-layer perceptron to encode each component in the formula, and then obtains the implicit features of the formula; The feature extraction subnetwork uses the self-attention mechanism to perform global correlation learning on implicit features to obtain deep features of the samples, and the prediction subnetwork learns the prediction functions of various performances from the deep features through a linear layer.
2. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 1, characterized in that: The molecular structure of biodegradable polyester includes chain structure and condensed structure, wherein the chain structure includes chemical composition, including main chain chemical composition and terminal group chemical composition; configuration, including optical isomerism; sequence structure and average composition; molecular size, including molecular weight and molecular weight distribution; The condensed structure includes: crystalline structure, including crystallinity, grain size, and lamellar thickness; orientation structure and orientation degree.
3. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 1, characterized in that: The molecular structure of the modification additive includes chemical composition, configuration, structure and molecular size.
4. The method for predicting the performance of biodegradable mulch films based on formula genes according to claim 1, characterized in that: The properties of ground film materials include thermodynamic properties, including glass transition temperature, melting temperature, crystallization temperature; rheology, melt index, color value, density, acid value, terminal hydroxyl content, thickness, light transmittance and UV shielding properties, mechanical properties, including tensile strength, elongation at break, flexural strength, flexural modulus, impact strength; water vapor and oxygen permeability.
5. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 1, characterized in that: Data preprocessing includes statistical analysis of the formulas of PBAT, PBST, PBSA, PLA, PPC, PPCP, PGA, PBS, PHA, PCL biodegradable materials and their modified additives, using the autoencoder method to project the performance of different formulas into a high-dimensional space, and then calculating the Euclidean distance between different formulas. The different formulas are clustered based on the distance matrix using the local outlier factor LOF method, and then all formula samples with a distance exceeding 50% of the longest distance are screened out, so that the distribution of formula samples covers all materials and the distribution of different formula data conforms to the average distribution.
6. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 1, characterized in that: The coding network encodes the molecular structure data in the biodegradable mulch film raw material information. On the one hand, the SMILES code of the molecular structure is converted into an extended connection fingerprint code through hash coding; on the other hand, the molecular structure graph is converted into a molecular structure feature value using graph theory, and further converted into a graph vector code using a graph convolution network; the extended connection fingerprint code and the graph vector code are spliced to obtain a combined code of the material; The molecular structure characteristic values include atomic vectors, bond vectors, and global vectors.
7. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 6, characterized in that: The molecular structure graph is converted into molecular structure feature values by using graph theory, and further converted into graph vector encoding by using graph convolution network as follows: Graph neural networks are used to process molecular structures. Each molecule is represented as a graph, where atoms are nodes and chemical bonds are edges. Node features are updated through graph convolution layers. The graph convolution formula is as follows: In the formula, X is the node feature, and A is the adjacency matrix. I is the identity matrix, is the degree matrix, W is the weight matrix, σ is the nonlinear activation function, and Y is the deep feature of X.
8. The method for predicting the performance of biodegradable mulch films based on recipe genes according to claim 6, characterized in that: The extended connection fingerprint code and the graph vector code are concatenated to obtain the combined code of the material. The feature concatenation is expressed as: X final =Concat(X Hash ,X GNN ) Where, X final is the final result of the combined encoding, X Hash is the result of hash encoding, X GNN The graph vector result obtained by the graph convolutional network.
9. A computing device, characterized in that include: One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the methods according to claims 1 to 8.
10. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions which, when executed by a computing device, cause the computing device to perform any one of the methods according to claims 1 to 8.