Experimental scheme design system combined with intelligent model
Through the experimental solution design system combined with intelligent models, the complex and time-consuming problem of polymer material research and development process is solved, and efficient experimental solution design and cost control are achieved.
Patent Information
- Application Number
- CN202510489329.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-18
- Publication Date
- 2025-07-18
AI Technical Summary
The research and development process of polymer materials is complex and time-consuming, and the design of traditional experimental solutions is inefficient and costly.
The experimental scheme design system combined with intelligent models includes the design main control module, monomer database, polymer database, property prediction model and model fine-tuning module. The experimental scheme design is carried out through intelligent models to reduce error rates and improve design efficiency.
It reduces the design error rate, improves the design efficiency, and can optimize the scheme according to the degree of synthesis.
Smart Images

Figure CN120340648A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and particularly relates to an experimental scheme design system combined with an intelligent model. Background Art
[0002] In the field of materials science, polymer materials play a crucial role in multiple industrial fields due to their unique physical and chemical properties. However, the R & D process of polymer materials is often complex and time-consuming, especially in the experimental scheme design stage. Traditional methods rely on a large number of trial-and-error experiments, which are not only inefficient but also costly. Summary of the Invention
[0003] The purpose of the present invention is to provide an experimental scheme design system combined with an intelligent model for the defects of the prior art. The system includes: a design main control module, a monomer database, a polymer database, a property prediction model, and a model fine-tuning module. Among them, the design main control module extracts the corresponding first polymer name and the first property index set from the polymer design task input by the user, obtains the first monomer SMILES sequence corresponding to the first polymer name by querying the polymer monomer dictionary, and uses this monomer SMILES sequence to query the monomer database to obtain a series of monomer information (i.e., the first monomer record set) with similar structures. Then, an experimental scheme design is carried out according to the polymer database, the polymer name, the first monomer record set, the property prediction model, and the first property index set to obtain the corresponding first experimental scheme sequence and feedback it to the current user. The monomer database is used to store multiple first monomer records, and each first monomer record corresponds to a monomer molecule. The polymer database is used to store multiple first polymer records, and each first polymer record corresponds to a polymer macromolecule. The property prediction model is used to perform corresponding multi-class physical property prediction and synthesizability prediction processing according to the first molecular graph sequence, the first ratio vector, and the first condition vector input into the model. The model fine-tuning module is used to perform a model fine-tuning process on the property prediction model based on the current polymer database before the property prediction model is used for the first time or after the polymer database is batch updated by the administrator each time. Through the present invention, not only can the error rate be reduced, but also the design efficiency can be improved, and the optimization can be carried out according to the predicted synthesizability.
[0004] To achieve the above object, an embodiment of the present invention provides an experimental scheme design system combined with an intelligent model, and the system includes: a design main control module, a monomer database, a polymer database, a property prediction model, and a model fine-tuning module;
[0005] The design main control module is respectively connected to the monomer database, the polymer database, and the property prediction model; the model fine-tuning module is respectively connected to the polymer database and the property prediction model;
[0006] The design master module is used to receive a polymer design task input by a user; extract the corresponding first polymer name and the first property index set from the polymer design task; query a preset polymer monomer dictionary according to the first polymer name to obtain the corresponding first monomer SMILES sequence; query the monomer database according to the first monomer SMILES sequence to obtain the corresponding first monomer record set; and design an experimental plan according to the polymer database, the first polymer name, the first monomer record set, the property prediction model, and the first property index set to obtain the corresponding first experimental plan sequence and feedback it to the current user.
[0007] The monomer database is used to store a plurality of first monomer records; each of the first monomer records corresponds to a monomer molecule.
[0008] The polymer database is used to store a plurality of first polymer records; each of the first polymer records corresponds to a polymer macromolecule; the number of the first polymer records corresponding to the same polymer macromolecule in the polymer database is not unique.
[0009] The polymer database is also used to store a plurality of the first polymer records regularly input by an administrator into the database.
[0010] The property prediction model is used to perform corresponding N-class physical property prediction and synthesizability prediction processing according to the input first molecular graph sequence, first ratio vector, and first condition vector to obtain the corresponding N physical property prediction values and one synthesizability prediction value; each of the physical property prediction values corresponds to a polymer physical property; N is a preset positive integer; the first molecular graph sequence is composed of a plurality of first molecular graphs; the first ratio vector is composed of a plurality of first ratios; the first ratio corresponds one-to-one with the first molecule; the first condition vector is composed of a polymerization reaction type code, a reaction temperature code, a reaction pressure code, a light type code, and a light intensity code. Figure 1 The first condition vector is composed of a polymerization reaction type code, a reaction temperature code, a reaction pressure code, a light type code, and a light intensity code.
[0011] The model fine-tuning module is used to perform a model fine-tuning process on the property prediction model based on the current polymer database before the property prediction model is used for the first time or after the polymer database is batch-updated by the administrator each time.
[0012] Preferably, the polymer design task includes the first polymer name and the first property index set; the first property index set is composed of one or more first property indexes; the first property index includes a first property type and a first property value range.
[0013] The polymer monomer dictionary includes a plurality of dictionary elements; each of the dictionary elements consists of a corresponding polymer name and a monomer SMILES sequence; all the polymer names in the polymer monomer dictionary are unique;
[0014] The first monomer record includes a first identifier, a first name, a first chemical formula, a first SMILES sequence, and a first molecular graph; the first identifier is the unique identifier of the current monomer record; the first name is the name of the current monomer molecule; the first chemical formula is the chemical formula of the current monomer molecule; the first SMILES sequence is the one-dimensional molecular sequence of the current monomer molecule; the first molecular graph is the three-dimensional molecular graph of the current monomer record; the three-dimensional molecular graph consists of a corresponding atom set and a chemical bond set; the atom set consists of a plurality of first atoms; the atomic attributes of the first atoms include at least an atom identifier, an atomic element type, and three-dimensional atomic coordinates; the chemical bond set consists of a plurality of first chemical bonds; the bond attributes of the first chemical bonds include at least a chemical bond identifier, a chemical bond type, and a chemical bond identifier group, and the chemical bond identifier group consists of the first atom identifiers of the two first atoms connected by the current chemical bond;
[0015] The first polymer record includes a second identifier, a second name, a first reaction component set, a first component ratio, a first reaction condition set, a first polymer property set, a first synthesizability score, and a first preparation process description; the second identifier is the unique identifier of the current polymer record; the second name is the name of the current polymer macromolecule; the first reaction component set consists of a plurality of first reaction components; the first reaction components include a component name, a component type, and a component molecular graph; the component type includes at least monomer, initiator, catalyst, and solvent; the component molecular graph is the three-dimensional molecular graph of the corresponding reaction component; the first component ratio is the molar ratio of all components of the current polymer macromolecule; the first reaction condition set includes a first polymerization reaction type, a first reaction temperature, a first reaction pressure, a first light type, and a first light intensity; the first polymerization reaction type includes at least free radical polymerization, anionic polymerization, cationic polymerization, and polycondensation reaction; the first light type includes at least ultraviolet light and visible light; the first polymer property set includes N first physical property scalars, and each of the first physical property scalars corresponds to a polymer physical property; the first synthesizability score is the synthesizability score of the current polymer macromolecule; the first preparation process description is the laboratory preparation process of the current polymer macromolecule.
[0016] Preferably, the design master module is specifically configured to, when querying the preset polymer monomer dictionary according to the first polymer name to obtain the corresponding first monomer SMILES sequence, use the monomer SMILES sequence of the dictionary element in the polymer monomer dictionary that matches the polymer name with the first polymer name as the corresponding first monomer SMILES sequence.
[0017] Preferably, the design master module is specifically configured to, when querying the monomer database according to the first monomer SMILES sequence to obtain the corresponding first monomer record set, perform molecular fingerprint conversion processing on the first monomer SMILES sequence to obtain the corresponding first molecular fingerprint; and perform molecular fingerprint conversion processing on the first SMILES sequence of each of the first monomer records in the monomer database to obtain the corresponding second molecular fingerprint; and calculate the similarity between the first molecular fingerprint and each of the second molecular fingerprints to obtain the corresponding first similarity; and extract the first monomer records corresponding to the first similarity that exceeds the preset first similarity threshold to form the corresponding first monomer record set.
[0018] Preferably, when the design master module feeds back the corresponding first experimental plan sequence to the current user according to the polymer database, the first polymer name, the first monomer record set, the property prediction model, and the first property index set during the experimental plan design:
[0019] Step 51, extract the first polymer records in the polymer database that match the second name with the first polymer name to form the corresponding first record set;
[0020] Step 52, perform a round of traversal on all the first polymer records in the first record set; during this round of traversal, take the currently traversed first polymer record as the corresponding current polymer record; perform molecular fingerprint conversion on the component molecular graph of the first reaction component with the component type of monomer in the current polymer record to obtain the corresponding third molecular fingerprint; perform molecular fingerprint conversion on each of the first molecular graphs in the first monomer record set to obtain the corresponding fourth molecular fingerprint; calculate the similarity between the third molecular fingerprint and each of the fourth molecular fingerprints to obtain the corresponding second similarity; identify whether all the second similarities corresponding to the current polymer record are lower than a preset second similarity threshold; if so, remove the current polymer record from the first record set; if not, record the first monomer records with the second similarity not lower than the second similarity threshold as the corresponding first records, and replace the component name and the component molecular graph of the first reaction component with the component type of monomer in the current polymer record with the first name and the first molecular graph of each of the first records to obtain one or more first replacement records, and reset the first polymer property set and the first synthesizability score of each of the first replacement records based on the property prediction model, and identify whether the first polymer property set of each reset first replacement record meets the first property index set, and if not, further delete the current first replacement record;
[0021] Step 53, after the end of this round of traversal, take each remaining first replacement record as a corresponding first experimental plan; sort all the obtained first experimental plans in descending order according to the first synthesizability score to form the corresponding first experimental plan sequence and feedback it to the current user.
[0022] Preferably, when the design master module specifically resets the first polymer property set and the first synthesizability score of each of the first replacement records based on the property prediction model:
[0023] Step 61, extract the first reaction component set, the first component ratio, and the first reaction condition set of the current first replacement record as the corresponding current reaction component set, current component ratio, and current reaction condition set;
[0024] Step 62, extract the component molecular graph of each of the first reaction components in the current reaction component set as the corresponding first molecular graph, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs;
[0025] Step 63, and use each proportion ratio of the current component ratio as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector;
[0026] Step 64, extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity from the current reaction condition set as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; perform one-hot encoding on the current polymerization reaction type to obtain the corresponding polymerization reaction type encoding, and perform one-hot encoding on the current light type to obtain the corresponding light intensity encoding; perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; and form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding;
[0027] Step 65, input the first molecular graph sequence, the first ratio vector, and the first condition vector obtained this time into the property prediction model to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain corresponding N physical property prediction quantities and one synthesizability prediction quantity;
[0028] Step 66, reset the first physical property scalar in the first polymer property set of the current first replacement record based on each physical property prediction quantity obtained this time; and reset the first synthesizability score of the current first replacement record based on the synthesizability prediction quantity obtained this time.
[0029] Preferably, the property prediction model includes a Uni-Mol model, a Uni-ELF backbone, N physical property prediction networks, and one synthesizability prediction network; each of the physical property prediction networks or the synthesizability prediction network is implemented based on a type of non-linear regression model, and the non-linear regression model at least includes an XGBoost model, a GBDT model, a Random Forest model, and an MLP model; the Uni-Mol model and the Uni-ELF backbone have both been pre-trained in advance;
[0030] The property prediction model is specifically used for performing corresponding N-class physical property prediction and synthesizability prediction processing on the first molecular graph sequence, the first ratio vector, and the first condition vector input according to the model to obtain corresponding N physical property prediction quantities and one synthesizability prediction quantity: automatically allocate a corresponding Uni-Mol model for each of the first molecular graphs in the first molecular graph sequence; perform atomic-level feature encoding processing on the corresponding first molecular graph by each of the Uni-Mol models to obtain corresponding first feature tensors; input all the obtained first feature tensors, the first ratio vector, and the first condition vector input by the model into the Uni-ELF backbone for solution-level feature encoding processing to obtain corresponding second feature tensors; perform corresponding physical property prediction processing on the second feature tensors by each of the physical property prediction networks and output corresponding physical property prediction quantities; and perform corresponding synthesizability prediction processing on the second feature tensors by the synthesizability prediction network and output corresponding synthesizability prediction quantities.
[0031] Preferably, the model fine-tuning module is specifically used for performing a first model fine-tuning process on the property prediction model based on the current polymer database:
[0032] Step 81, extract a plurality of the first polymer records in the polymer database whose update time is close to the current time according to a preset first ratio to form a corresponding first data set; divide the first data set into two sub-data sets according to a preset second ratio and denote them as corresponding first training set and first evaluation set; and use the first polymer record of the first training set as the corresponding current training record;
[0033] Among them, the first data set, the first training set, and the first evaluation set are all composed of a plurality of the first polymer records; the ratio of the total number of records in the first data set to the total number of records in the polymer database satisfies the first ratio; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the second ratio;
[0034] Step 82: Extract each component molecular graph in the first reaction component set of the current training record as the corresponding first molecular graph, and form the corresponding first molecular graph sequence with all the obtained first molecular graphs; and take each proportion ratio in the first component ratio of the current training record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector; and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity in the first reaction condition set of the current training record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; and perform one-hot encoding on the current polymerization reaction type and the current light type respectively to obtain the corresponding polymerization reaction type encoding and light intensity encoding; and perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; and form the corresponding first condition vector with the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding.
[0035] Step 83: Input the obtained first molecular graph sequence, the first ratio vector, and the first condition vector into the property prediction model to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction values and one synthesizability prediction value; and form a first prediction vector with a length of N + 1 with the N physical property prediction values and the synthesizability prediction value obtained this time.
[0036] Step 84: Form a first label vector with a length of N + 1 with the N first physical property scalars and the first synthesizability score of the current training record; and input the first prediction vector and the first label vector into a preset first model loss function; and perform one round of fine-tuning on the model parameters of the property prediction model in the direction of minimizing the first model loss function based on a preset first model optimizer.
[0037] Among them, the first model loss function at least includes the L1 loss function, the L2 loss function, and the cross-entropy loss function; the first model optimizer at least includes the SGD optimizer and the Adam optimizer.
[0038] Step 85: Identify whether the current training record is the last first polymer record in the first training set; if so, go to Step 86; if not, extract the next first polymer record in the first training set as the new current training record and return to Step 82;
[0039] Step 86: Take the first first polymer record in the first evaluation set as the corresponding current evaluation record;
[0040] Step 87: Extract each component molecular graph in the first reaction component set of the current evaluation record as the corresponding first molecular graph, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs; take each proportion ratio of the first component ratio of the current evaluation record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector; extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity in the first reaction condition set of the current evaluation record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; perform one-hot encoding on the current polymerization reaction type and the current light type respectively to obtain the corresponding polymerization reaction type encoding and light intensity encoding; perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding; input the first molecular graph sequence, the first ratio vector, and the first condition vector obtained this time into the property prediction model for corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction values and one synthesizability prediction value; form a second prediction vector with a length of N + 1 from the N physical property prediction values and the synthesizability prediction value obtained this time; form a second label vector with a length of N + 1 from the N first physical property scalars and the first synthesizability score of the current evaluation record; form a corresponding first prediction-label pair from the second prediction vector and the second label vector obtained this time;
[0041] Step 88: Identify whether the current evaluation record is the last first polymer record in the first evaluation set; if so, go to Step 89; if not, extract the next first polymer record in the first evaluation set as the new current evaluation record and return to Step 87;
[0042] Step 89: Bring all the obtained first prediction-label pairs into a preset first model evaluation function for calculation to obtain corresponding first evaluation values; and identify whether the first evaluation values meet a preset first evaluation value range; if not, use the first first polymer record in the first training set as the new current training record and return to Step 82; if so, confirm that the current model fine-tuning is completed;
[0043] Among them, the first model evaluation function at least includes the RMSE function.
[0044] The embodiment of the present invention provides an experimental scheme design system combined with an intelligent model. The system includes: a design main control module, a monomer database, a polymer database, a property prediction model, and a model fine-tuning module. Among them, the design main control module extracts the corresponding first polymer name and the first property index set from the polymer design task input by the user, obtains the first monomer SMILES sequence corresponding to the first polymer name by querying the polymer monomer dictionary, and uses this monomer SMILES sequence to query the monomer database to obtain a series of monomer information with similar structures (i.e., the first monomer record set), and then designs an experimental scheme according to the polymer database, the polymer name, the first monomer record set, the property prediction model, and the first property index set to obtain a corresponding first experimental scheme sequence and feedback it to the current user. The monomer database is used to store multiple first monomer records, and each first monomer record corresponds to a monomer molecule. The polymer database is used to store multiple first polymer records, and each first polymer record corresponds to a polymer macromolecule. The property prediction model is used to perform corresponding multi-class physical property prediction and synthesizability prediction processing according to the first molecular graph sequence, the first ratio vector, and the first condition vector input to the model. The model fine-tuning module is used to perform a model fine-tuning process on the property prediction model based on the current polymer database before the property prediction model is used for the first time or after the polymer database is batch updated by the administrator each time. Through the embodiment of the present invention, not only the design error rate is reduced, but also the design efficiency is improved, and the scheme can be optimized according to the predicted synthesizability. Description of the Drawings
[0045] Figure 1 It is a module structure diagram of an experimental scheme design system combined with an intelligent model provided by the embodiment of the present invention;
[0046] Figure 2It is the module structure diagram of the property prediction model provided by the embodiment of the present invention. Detailed implementation manners
[0047] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0048] The experimental scheme design system combined with an intelligent model provided by the embodiment of the present invention, as Figure 1 shown in the module structure diagram of an experimental scheme design system combined with an intelligent model provided by the embodiment of the present invention, mainly includes: a design main control module 1, a monomer database 2, a polymer database 3, a property prediction model 4, and a model fine-tuning module 5.
[0049] The connection relationships of each system module are as follows: the design main control module 1 is respectively connected to the monomer database 2, the polymer database 3, and the property prediction model 4; the model fine-tuning module 5 is respectively connected to the polymer database 3 and the property prediction model 4.
[0050] (1) Design main control module 1:
[0051] The design main control module 1 of the embodiment of the present invention is used to receive a polymer design task input by a user; extract a corresponding first polymer name and a first property index set from the polymer design task; query a preset polymer monomer dictionary according to the first polymer name to obtain a corresponding first monomer SMILES sequence; query the monomer database 2 according to the first monomer SMILES sequence to obtain a corresponding first monomer record set; and perform experimental scheme design according to the polymer database 3, the first polymer name, the first monomer record set, the property prediction model 4, and the first property index set to obtain a corresponding first experimental scheme sequence and feedback it to the current user.
[0052] Here, the polymer design task includes a first polymer name and a first property index set; wherein, the first property index set is composed of one or more first property indexes; the first property index includes a first property type and a first property value range; the first property type includes multiple types of polymer physical properties.
[0053] The polymer monomer dictionary includes multiple dictionary elements; each dictionary element is composed of a corresponding polymer name and a monomer SMILES sequence; all polymer names in the polymer monomer dictionary are unique.
[0054] In a specific implementation manner of an embodiment of the present invention, the designed main control module 1 is specifically configured to, when querying a preset polymer monomer dictionary according to the first polymer name to obtain the corresponding first monomer SMILES sequence, use the monomer SMILES sequence of the dictionary element in the polymer monomer dictionary whose polymer name matches the first polymer name as the corresponding first monomer SMILES sequence.
[0055] In another specific implementation manner of an embodiment of the present invention, the designed main control module 1 is specifically configured to, when querying the monomer database 2 according to the first monomer SMILES sequence to obtain the corresponding first monomer record set, perform molecular fingerprint conversion processing on the first monomer SMILES sequence to obtain the corresponding first molecular fingerprint; perform molecular fingerprint conversion processing on the first SMILES sequence of each first monomer record in the monomer database 2 to obtain the corresponding second molecular fingerprint; calculate the similarity between the first molecular fingerprint and each second molecular fingerprint to obtain the corresponding first similarity; and extract the first monomer records corresponding to the first similarities that exceed the preset first similarity threshold to form the corresponding first monomer record set. Here, the first similarity threshold is a preset threshold parameter.
[0056] In another specific implementation manner of an embodiment of the present invention, the designed main control module 1 is specifically configured to, when feeding back the corresponding first experimental plan sequence to the current user according to the polymer database 3, the first polymer name, the first monomer record set, the property prediction model 4, and the first property index set for experimental plan design:
[0057] Step A1, extract the first polymer records in the polymer database 3 whose second name matches the first polymer name to form the corresponding first record set;
[0058] Step A2, perform a round of traversal on all the first polymer records in the first record set; during this round of traversal, take the currently traversed first polymer record as the corresponding current polymer record; perform molecular fingerprint conversion on the component molecular graph of the first reaction component with the monomer component type in the current polymer record to obtain the corresponding third molecular fingerprint; perform molecular fingerprint conversion on each first molecular graph in the first monomer record set to obtain the corresponding fourth molecular fingerprint; calculate the similarity between the third molecular fingerprint and each fourth molecular fingerprint to obtain the corresponding second similarity; identify whether all the second similarities corresponding to the current polymer record are lower than the preset second similarity threshold; if so, remove the current polymer record from the first record set; if not, record the first monomer records with second similarities not lower than the second similarity threshold as the corresponding first records, and replace the component name and component molecular graph of the first reaction component with the monomer component type in the current polymer record with the first name and first molecular graph of each first record to obtain one or more first replacement records, and reset the first polymer property set and the first synthesizability score of each first replacement record based on the property prediction model 4, and identify whether the first polymer property set of each reset first replacement record meets the first property index set, and if not, further delete the current first replacement record;
[0059] Here, the second similarity threshold is a preset threshold parameter;
[0060] Step A3, after the end of this round of traversal, take each remaining first replacement record as a corresponding first experimental scheme; sort all the obtained first experimental schemes in descending order of the first synthesizability score to form the corresponding first experimental scheme sequence and feedback it to the current user.
[0061] In another specific implementation manner of the embodiment of the present invention, the design master control module 1 is specifically used for resetting the first polymer property set and the first synthesizability score of each first replacement record based on the property prediction model 4:
[0062] Step B1, extract the first reaction component set, the first component ratio, and the first reaction condition set of the current first replacement record as the corresponding current reaction component set, the current component ratio, and the current reaction condition set;
[0063] Step B2, extract the component molecular graph of each first reaction component in the current reaction component set as the corresponding first molecular graph, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs;
[0064] Step B3, and use the respective proportion ratios of the current composition ratio as the corresponding first ratios, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector;
[0065] Step B4, and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity of the current reaction condition set as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; and perform one-hot encoding on the current polymerization reaction type to obtain the corresponding polymerization reaction type encoding, and perform one-hot encoding on the current light type to obtain the corresponding light intensity encoding; and perform normalization processing on the current reaction temperature based on the preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on the preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light intensity based on the preset light intensity range to obtain the corresponding light intensity encoding; and form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding;
[0066] Here, the reaction temperature range, the reaction pressure range, and the light intensity range are three preset numerical ranges;
[0067] Step B5, input the first molecular graph sequence, the first ratio vector, and the first condition vector obtained this time into the property prediction model 4 to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction values and one synthesizability prediction value;
[0068] Step B6, reset the first physical property scalar in the first polymer property set of the current first replacement record based on each physical property prediction value obtained this time; and reset the first synthesizability score of the current first replacement record based on the synthesizability prediction value obtained this time.
[0069] (2) Monomer database 2:
[0070] The monomer database 2 of the embodiment of the present invention is used to store multiple first monomer records.
[0071] Here, each first monomer record corresponds to a monomer molecule. The first monomer record includes a first identifier, a first name, a first chemical formula, a first SMILES sequence, and a first molecular graph; the first identifier is the unique identifier of the current monomer record; the first name is the name of the current monomer molecule; the first chemical formula is the chemical formula of the current monomer molecule; the first SMILES sequence is the one-dimensional molecular sequence of the current monomer molecule; the first molecular graph is the three-dimensional molecular graph of the current monomer record. It should be noted that the three-dimensional molecular graph mentioned in the embodiments of the present invention is a data object of a graph structure, composed of a corresponding atom set and a chemical bond set; the atom set is composed of multiple first atoms; the atomic attributes of the first atoms at least include atomic identifiers, atomic element types, and three-dimensional atomic coordinates; the chemical bond set is composed of multiple first chemical bonds; the bond attributes of the first chemical bonds at least include chemical bond identifiers, chemical bond types, and chemical bond identifier groups, and the chemical bond identifier group is composed of the first atomic identifiers of the two first atoms connected by the current chemical bond.
[0072] (III) Polymer Database 3:
[0073] The polymer database 3 in the embodiments of the present invention is used to store multiple first polymer records. The polymer database 3 also stores the multiple first polymer records regularly entered by the administrator in the database.
[0074] Here, each first polymer record corresponds to a polymer macromolecule; the number of first polymer records corresponding to the same polymer macromolecule in the polymer database 3 is not unique. The first polymer record includes a second identifier, a second name, a first reaction component set, a first composition ratio, a first reaction condition set, a first polymer property set, a first synthesizability score, and a first preparation process description; the second identifier is the unique identifier of the current polymer record; the second name is the name of the current polymer macromolecule; the first reaction component set is composed of multiple first reaction components; the first reaction component includes a component name, a component type, and a component molecular graph; the component type at least includes monomers, initiators, catalysts, solvents; the component molecular graph is the three-dimensional molecular graph of the corresponding reaction component; the first composition ratio is the molar ratio of all components of the current polymer macromolecule; the first reaction condition set includes a first polymerization reaction type, a first reaction temperature, a first reaction pressure, a first light irradiation type, and a first light irradiation intensity; the first polymerization reaction type at least includes free radical polymerization, anionic polymerization, cationic polymerization, and polycondensation reaction; the first light irradiation type at least includes ultraviolet light and visible light; the first polymer property set includes N first physical property scalars, and each first physical property scalar corresponds to a polymer physical property; the first synthesizability score is the synthesizability score of the current polymer macromolecule; the first preparation process description is the laboratory preparation process of the current polymer macromolecule.
[0075] (IV) Property Prediction Model 4:
[0076] The property prediction model 4 of the embodiments of the present invention is used to perform corresponding N-class physical property prediction and synthesizability prediction processing according to the first molecular graph sequence, the first ratio vector, and the first condition vector input to the model to obtain corresponding N physical property prediction values and one synthesizability prediction value. Among them, each physical property prediction value corresponds to a polymer physical property; N is a preset positive integer; the first molecular graph sequence is composed of multiple first molecular graphs; the first ratio vector is composed of multiple first ratios; the first ratio corresponds to the first molecule Figure 1 in a one-to-one correspondence; the first condition vector is composed of a polymerization reaction type code, a reaction temperature code, a reaction pressure code, a light type code, and a light intensity code.
[0077] Here, as Figure 2 shown in the module structure diagram of the property prediction model provided by the embodiments of the present invention, the property prediction model 4 of the embodiments of the present invention includes a Uni-Mol model, a Uni-ELF backbone, N physical property prediction networks, and one synthesizability prediction network.
[0078] It should be noted that each of the physical property prediction networks and the synthesizability prediction network of the embodiments of the present invention is implemented based on a class of non-linear regression models. The non-linear regression models mentioned here at least include the XGBoost model, the GBDT model, the Random Forest model, and the MLP model.
[0079] It should also be noted that both the Uni-Mol model and the Uni-ELF backbone have been pre-trained. For the design and training technical details of the Uni-Mol model, reference can be made to the technical literature "Uni-Mol: A Universal 3D Molecular Representation Learning Framework" for understanding; in the embodiments of the present invention, the combination method of the Uni-Mol model and the Uni-ELF backbone is implemented by referring to the two-stage framework of Uni-ELF. The specific technical content can be referred to the technical literature "Uni-ELF: A Multi-Level Representation Learning Framework for Electrolyte Formulation Design" for understanding.
[0080] In another specific implementation manner of the embodiment of the present invention, the property prediction model 4 is specifically used for, when performing corresponding N-class physical property prediction and synthesizability prediction processing on the basis of the first molecular graph sequence, the first ratio vector, and the first condition vector input to the model to obtain corresponding N physical property prediction quantities and one synthesizability prediction quantity: automatically allocating a corresponding Uni-Mol model to each first molecular graph of the first molecular graph sequence; performing atomic-level feature encoding processing on the corresponding first molecular graph by each Uni-Mol model to obtain corresponding first feature tensors; inputting all the obtained first feature tensors, the first ratio vector, and the first condition vector input to the model into the Uni-ELF backbone for solution-level feature encoding processing to obtain corresponding second feature tensors; performing corresponding physical property prediction processing by each physical property prediction network according to the second feature tensors and outputting corresponding physical property prediction quantities; and performing corresponding synthesizability prediction processing by the synthesizability prediction network according to the second feature tensors and outputting corresponding synthesizability prediction quantities.
[0081] (V) Model fine-tuning module 5:
[0082] The model fine-tuning module 5 of the embodiment of the present invention is used to perform a model fine-tuning process on the property prediction model 4 based on the current polymer database 3 before the property prediction model 4 is used for the first time or after the polymer database 3 is batch-updated by the administrator each time.
[0083] In another specific implementation manner of the embodiment of the present invention, the model fine-tuning module 5 is specifically used for, when performing a model fine-tuning process on the property prediction model 4 based on the current polymer database 3:
[0084] Step 81, extracting a plurality of first polymer records in the polymer database 3 whose update time is close to the current time according to a preset first ratio to form a corresponding first data set; splitting the first data set into two sub-data sets according to a preset second ratio and recording them as corresponding first training set and first evaluation set; and using the first first polymer record in the first training set as the corresponding current training record.
[0085] Wherein, the first and second ratios are two preset ratio parameters, such as 1 / 3, 8:2; the first data set, the first training set, and the first evaluation set are all composed of a plurality of first polymer records; the ratio of the total number of records in the first data set to the total number of records in the polymer database 3 satisfies the first ratio; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the second ratio.
[0086] Step 82: Extract each component molecular graph from the first reaction component set of the current training record as the corresponding first molecular graph, and form a corresponding first molecular graph sequence with all the obtained first molecular graphs; and take each proportion ratio of the first component ratio of the current training record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form a corresponding first proportion vector; and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity from the first reaction condition set of the current training record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; and perform one-hot encoding on the current polymerization reaction type and the current light type respectively to obtain the corresponding polymerization reaction type encoding and light intensity encoding; and perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; and form a corresponding first condition vector with the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding.
[0087] Step 83: Input the first molecular graph sequence, the first proportion vector, and the first condition vector obtained this time into the property prediction model 4 to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain corresponding N physical property prediction values and one synthesizability prediction value; and form a first prediction vector with a length of N + 1 with the N physical property prediction values and the synthesizability prediction value obtained this time.
[0088] Step 84: Form a first label vector with a length of N + 1 with the N first physical property scalars and the first synthesizability score of the current training record; and input the first prediction vector and the first label vector into a preset first model loss function; and perform one round of fine-tuning on the model parameters of the property prediction model 4 in the direction of minimizing the first model loss function based on a preset first model optimizer.
[0089] Among them, the first model loss function at least includes the L1 loss function, the L2 loss function, and the cross-entropy loss function; the first model optimizer at least includes the SGD optimizer and the Adam optimizer.
[0090] Step 85: Identify whether the current training record is the last first polymer record of the first training set; if so, go to Step 86; if not, extract the next first polymer record of the first training set as the new current training record and return to Step 82.
[0091] Step 86: Take the first first-polymer record of the first evaluation set as the corresponding current evaluation record;
[0092] Step 87: Extract each component molecular graph in the first reaction component set of the current evaluation record as the corresponding first molecular graph, and form a corresponding first molecular graph sequence from all the obtained first molecular graphs; and take each proportion ratio in the first component ratio of the current evaluation record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form a corresponding first proportion vector; and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity in the first reaction condition set of the current evaluation record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; and perform one-hot encoding on the current polymerization reaction type and the current light type respectively to obtain the corresponding polymerization reaction type encoding and light intensity encoding; and perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; and form a corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding; and input the first molecular graph sequence, the first proportion vector, and the first condition vector obtained this time into the property prediction model 4 for corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction values and one synthesizability prediction value; and form a second prediction vector with a length of N + 1 from the N physical property prediction values and the synthesizability prediction value obtained this time; and form a second label vector with a length of N + 1 from the N first physical property scalars and the first synthesizability score of the current evaluation record; and form a corresponding first prediction-label pair from the second prediction vector and the second label vector obtained this time;
[0093] Step 88: Identify whether the current evaluation record is the last first-polymer record of the first evaluation set; if so, go to Step 89; if not, extract the next first-polymer record of the first evaluation set as the new current evaluation record and return to Step 87;
[0094] Step 89: Substitute all the obtained first prediction-label pairs into a preset first model evaluation function for calculation to obtain the corresponding first evaluation value; and identify whether the first evaluation value meets a preset first evaluation value range; if not, take the first first-polymer record of the first training set as the new current training record and return to Step 82; if so, confirm that the current model fine-tuning is completed;
[0095] Among them, the first evaluation value range is a pre-set numerical range; the first model evaluation function includes at least the RMSE function.
[0096] It should also be noted that it should be understood that the division of each module of the above system is only a division of logical functions. In actual implementation, it can be fully or partially integrated into a physical entity, or physically separated. And these modules can all be implemented in the form of software called by a processing element; they can also all be implemented in the form of hardware; or some modules can be implemented in the form of software called by a processing element, and some modules can be implemented in the form of hardware. For example, the design master control module can be a separately established processing element, or can be integrated in a certain chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a certain processing element of the above system to perform the functions of the above determined module. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together or can be independently implemented. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each method step of the foregoing method or each module processing step of the foregoing system can be completed by the integrated logic circuit in the hardware of the processor element or the instruction in the form of software.
[0097] For example, these modules of the above system can be one or more integrated circuits configured to implement the foregoing method, such as: one or more Application Specific Integrated Circuits (ASICs), or, one or more Digital Signal Processors (DSPs), or, one or more Field Programmable Gate Arrays (FPGAs), etc. Again, when a certain module of the above system is implemented in the form of a processing element scheduling program code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processors that can call program code. Again, these modules can be integrated together and implemented in the form of a System-on-a-chip (SOC).
[0098] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the foregoing method embodiments are generated in whole or in part. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the above computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.). The above computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The above available medium may be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.
[0099] An embodiment of the present invention provides an experimental scheme design system combined with an intelligent model. The system includes: a design main control module, a monomer database, a polymer database, a property prediction model, and a model fine-tuning module. Among them, the design main control module extracts the corresponding first polymer name and the first property index set from the polymer design task input by the user, obtains the first monomer SMILES sequence corresponding to the first polymer name by querying the polymer monomer dictionary, and uses the monomer SMILES sequence to query the monomer database to obtain a series of monomer information with similar structures (i.e., the first monomer record set). Then, an experimental scheme design is carried out according to the polymer database, the polymer name, the first monomer record set, the property prediction model, and the first property index set to obtain the corresponding first experimental scheme sequence and feedback it to the current user. The monomer database is used to store multiple first monomer records, and each first monomer record corresponds to a monomer molecule. The polymer database is used to store multiple first polymer records, and each first polymer record corresponds to a polymer macromolecule. The property prediction model is used to perform corresponding multi-class physical property prediction and synthesizability prediction processing according to the first molecular graph sequence, the first ratio vector, and the first condition vector input into the model. The model fine-tuning module is used to perform a model fine-tuning process on the property prediction model based on the current polymer database before the property prediction model is used for the first time or after the polymer database is batch updated by the administrator each time. Through the embodiment of the present invention, not only the design error rate is reduced, but also the design efficiency is improved, and the scheme can be optimized according to the predicted synthesizability.
[0100] Those skilled in the art should also be able to further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.
[0101] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.
[0102] The specific embodiments described above further elaborate on the object, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims
1. An experimental scheme design system combined with an intelligent model, characterized in that The system includes: a design master control module, a monomer database, a polymer database, a property prediction model, and a model fine-tuning module; The design master control module is respectively connected to the monomer database, the polymer database, and the property prediction model; the model fine-tuning module is respectively connected to the polymer database and the property prediction model; The design master control module is configured to receive a polymer design task input by a user; extract a corresponding first polymer name and a first property index set from the polymer design task; query a preset polymer monomer dictionary according to the first polymer name to obtain a corresponding first monomer SMILES sequence; query the monomer database according to the first monomer SMILES sequence to obtain a corresponding first monomer record set; and perform experimental scheme design according to the polymer database, the first polymer name, the first monomer record set, the property prediction model, and the first property index set to obtain a corresponding first experimental scheme sequence and feedback it to the current user; The monomer database is used to store a plurality of first monomer records; each of the first monomer records corresponds to a monomer molecule; The polymer database is used to store a plurality of first polymer records; each of the first polymer records corresponds to a polymer macromolecule; the number of the first polymer records corresponding to the same polymer macromolecule in the polymer database is not unique; The polymer database is also used to store a plurality of the first polymer records regularly input by an administrator in the database; The property prediction model is configured to perform corresponding N types of physical property prediction and synthesizability prediction processing according to a first molecular graph sequence, a first ratio vector, and a first condition vector input to the model to obtain corresponding N physical property prediction values and a synthesizability prediction value; each of the physical property prediction values corresponds to a polymer physical property; N is a preset positive integer; the first molecular graph sequence is composed of a plurality of first molecular graphs; the first ratio vector is composed of a plurality of first ratios; the first ratios correspond to the first molecular graphs one by one; the first condition vector is composed of a polymerization reaction type code, a reaction temperature code, a reaction pressure code, a light type code, and a light intensity code; The model fine-tuning module is configured to perform a model fine-tuning process on the property prediction model based on the current polymer database before the property prediction model is used for the first time or after the polymer database is batch-updated by the administrator each time.
2. The experimental scheme design system combined with an intelligent model according to claim 1, wherein The polymer design task includes the first polymer name and the first property index set; the first property index set is composed of one or more first property indexes; the first property index includes a first property type and a first property value range; The polymer monomer dictionary includes a plurality of dictionary elements; each of the dictionary elements is composed of a corresponding polymer name and a monomer SMILES sequence; all the polymer names in the polymer monomer dictionary are unique; The first monomer record includes a first identifier, a first name, a first chemical formula, a first SMILES sequence, and a first molecular graph; the first identifier is the unique identifier of the current monomer record; the first name is the name of the current monomer molecule; the first chemical formula is the chemical formula of the current monomer molecule; the first SMILES sequence is the one-dimensional molecular sequence of the current monomer molecule; The first molecular graph is the three-dimensional molecular graph of the current monomer record; the three-dimensional molecular graph is composed of a corresponding atom set and a chemical bond set; the atom set is composed of multiple first atoms; the atomic attributes of the first atoms at least include atom identifiers, atomic element types, and three-dimensional atomic coordinates; the chemical bond set is composed of multiple first chemical bonds; the bond attributes of the first chemical bonds at least include chemical bond identifiers, chemical bond types, and chemical bond identifier groups, and the chemical bond identifier groups are composed of the first atom identifiers of the two first atoms connected by the current chemical bond; The first polymer record includes a second identifier, a second name, a first reaction component set, a first composition ratio, a first reaction condition set, a first polymer property set, a first synthesizability score, and a first preparation process description; the second identifier is the unique identifier of the current polymer record; the second name is the name of the current polymer; the first reaction component set is composed of multiple first reaction components; the first reaction components include component names, component types, and component molecular graphs; the component types at least include monomers, initiators, catalysts, and solvents; the component molecular graph is the three-dimensional molecular graph of the corresponding reaction component; the first composition ratio is the molar ratio of all components of the current polymer; the first reaction condition set includes a first polymerization reaction type, a first reaction temperature, a first reaction pressure, a first light type, and a first light intensity; the first polymerization reaction type at least includes radical polymerization, anionic polymerization, cationic polymerization, and polycondensation reaction; the first light type at least includes ultraviolet light and visible light; the first polymer property set includes N first physical property scalars, and each first physical property scalar corresponds to a polymer physical property; the first synthesizability score is the synthesizability score of the current polymer; the first preparation process description is the laboratory preparation process of the current polymer.
3. The experimental scheme design system combined with an intelligent model according to claim 2, wherein The design main control module is specifically configured to use the monomer SMILES sequence of the dictionary element in the polymer monomer dictionary that matches the polymer name as the corresponding first monomer SMILES sequence when querying the preset polymer monomer dictionary according to the first polymer name.
4. The experimental scheme design system combined with an intelligent model according to claim 2, wherein The design master control module is specifically configured to, when querying the monomer database according to the first monomer SMILES sequence to obtain a corresponding first monomer record set, perform molecular fingerprint conversion processing on the first monomer SMILES sequence to obtain a corresponding first molecular fingerprint; and perform molecular fingerprint conversion processing on the first SMILES sequences of each of the first monomer records in the monomer database to obtain corresponding second molecular fingerprints; calculate the similarity between the first molecular fingerprint and each of the second molecular fingerprints to obtain corresponding first similarities; and extract the first monomer records corresponding to the first similarities that exceed a preset first similarity threshold to form the corresponding first monomer record set.
5. The experimental scheme design system combined with an intelligent model according to claim 2, wherein the design master control module is specifically configured to, when feeding back a corresponding first experimental scheme sequence to the current user according to the polymer database, the first polymer name, the first monomer record set, the property prediction model, and the first property index set: Step 51, extract the first polymer records in the polymer database whose second name matches the first polymer name to form a corresponding first record set; Step 52, and perform a round of traversal on all the first polymer records in the first record set; during this round of traversal, regard the currently traversed first polymer record as the corresponding current polymer record; perform molecular fingerprint conversion on the component molecular graph of the first reaction component whose component type is monomer in the current polymer record to obtain a corresponding third molecular fingerprint; perform molecular fingerprint conversion on each of the first molecular graphs in the first monomer record set to obtain corresponding fourth molecular fingerprints; calculate the similarity between the third molecular fingerprint and each of the fourth molecular fingerprints to obtain corresponding second similarities; and identify whether all the second similarities corresponding to the current polymer record are lower than a preset second similarity threshold; if so, remove the current polymer record from the first record set; if not, record the first monomer records whose second similarities are not lower than the second similarity threshold as the corresponding first records, and replace the component name and the component molecular graph of the first reaction component whose component type is monomer in the current polymer record with the first name and the first molecular graph of each of the first records to obtain one or more first replacement records, and reset the first polymer property set and the first synthesizability score of each of the first replacement records based on the property prediction model, and identify whether the first polymer property set of each reset first replacement record meets the first property index set, and if not, further delete the current first replacement record; Step 53, after the end of this round of traversal, take each remaining first replacement record as a corresponding first experimental scheme; and sort all the obtained first experimental schemes in descending order according to the first synthesizability score to form the corresponding first experimental scheme sequence and feedback it to the current user.
6. The experimental scheme design system combined with an intelligent model according to claim 2, wherein when the design main control module specifically resets the first polymer property set and the first synthesizability score of each first replacement record based on the property prediction model: Step 61, extract the first reaction component set, the first composition ratio, and the first reaction condition set of the current first replacement record as the corresponding current reaction component set, current composition ratio, and current reaction condition set; Step 62, extract the component molecular graphs of each first reaction component in the current reaction component set as the corresponding first molecular graphs, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs; Step 63, take each ratio in the current composition ratio as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector; Step 64, extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity of the current reaction condition set as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; perform one-hot encoding on the current polymerization reaction type to obtain the corresponding polymerization reaction type encoding, and perform one-hot encoding on the current light type to obtain the corresponding light intensity encoding; Perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; Perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; And form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding; Step 65, input the first molecular graph sequence, the first ratio vector, and the first condition vector obtained this time into the property prediction model to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction quantities and one synthesizability prediction quantity; Step 66, reset the first physical property scalar process corresponding to the first polymer property set in the current first replacement record based on each of the obtained physical property prediction values this time; and reset the first synthesizability score in the current first replacement record based on the obtained synthesizability prediction value this time.
7. The experimental scheme design system combined with an intelligent model according to claim 2, wherein the property prediction model includes a Uni-Mol model, a Uni-ELF backbone, N physical property prediction networks, and a synthesizability prediction network; each of the physical property prediction networks or the synthesizability prediction network is implemented based on a type of non-linear regression model, and the non-linear regression model at least includes an XGBoost model, a GBDT model, a Random Forest model, and an MLP model; the Uni-Mol model and the Uni-ELF backbone have both been pre-trained. The property prediction model is specifically used for performing corresponding N types of physical property prediction and synthesizability prediction processing on the first molecular graph sequence, the first ratio vector, and the first condition vector input according to the model to obtain corresponding N physical property prediction values and a synthesizability prediction value: automatically assign a corresponding Uni-Mol model to each of the first molecular graphs in the first molecular graph sequence; and perform atomic-level feature encoding processing on the corresponding first molecular graph by each of the Uni-Mol models to obtain corresponding first feature tensors; and input all the obtained first feature tensors, the first ratio vector, and the first condition vector input by the model into the Uni-ELF backbone for solution-level feature encoding processing to obtain corresponding second feature tensors; and perform corresponding physical property prediction processing by each of the physical property prediction networks according to the second feature tensors and output the corresponding physical property prediction values; and perform corresponding synthesizability prediction processing by the synthesizability prediction network according to the second feature tensors and output the corresponding synthesizability prediction value.
8. The experimental scheme design system combined with an intelligent model according to claim 2, wherein the model fine-tuning module is specifically used for performing a first model fine-tuning process on the property prediction model based on the current polymer database: Step 81, extract a plurality of the first polymer records in the polymer database whose update time is close to the current time according to a preset first ratio to form a corresponding first data set; and divide the first data set into two sub-data sets according to a preset second ratio and denote them as corresponding first training set and first evaluation set; and use the first polymer record of the first training set as the corresponding current training record. Among them, the first data set, the first training set, and the first evaluation set are all composed of multiple first polymer records; the ratio of the total number of records in the first data set to the total number of records in the polymer database satisfies the first ratio; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the second ratio; Step 82: Extract each component molecular graph in the first reaction component set of the current training record as the corresponding first molecular graph, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs; and use each proportion ratio in the first component ratio of the current training record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first proportion vector; and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light illumination type, and the first light illumination intensity in the first reaction condition set of the current training record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light illumination type, and current light illumination intensity; and perform one-hot encoding on the current polymerization reaction type and the current light illumination type respectively to obtain the corresponding polymerization reaction type encoding and light illumination intensity encoding; and perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light illumination intensity based on a preset light illumination intensity range to obtain the corresponding light illumination intensity encoding; and form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light illumination type encoding, and light illumination intensity encoding; Step 83: Input the first molecular graph sequence, the first proportion vector, and the first condition vector obtained this time into the property prediction model to perform corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction quantities and one synthesizability prediction quantity; and form a first prediction vector with a length of N + 1 from the N physical property prediction quantities and the synthesizability prediction quantity obtained this time; Step 84: Form a first label vector with a length of N + 1 from the N first physical property scalars and the first synthesizability score of the current training record; and input the first prediction vector and the first label vector into a preset first model loss function; and perform one round of fine-tuning on the model parameters of the property prediction model based on a preset first model optimizer in the direction of minimizing the first model loss function; Among them, the first model loss function at least includes the L1 loss function, the L2 loss function, and the cross-entropy loss function; the first model optimizer at least includes the SGD optimizer and the Adam optimizer; Step 85, identify whether the current training record is the last first polymer record in the first training set; if so, go to Step 86; if not, extract the next first polymer record in the first training set as the new current training record and return to Step 82; Step 86, take the first first polymer record in the first evaluation set as the corresponding current evaluation record; Step 87, extract each component molecular graph in the first reaction component set of the current evaluation record as the corresponding first molecular graph, and form the corresponding first molecular graph sequence from all the obtained first molecular graphs; and take each proportion ratio of the first component ratio of the current evaluation record as the corresponding first ratio, and sort all the obtained first ratios in the sorting order corresponding to the first molecular graph sequence to form the corresponding first ratio vector; and extract the first polymerization reaction type, the first reaction temperature, the first reaction pressure, the first light type, and the first light intensity in the first reaction condition set of the current evaluation record as the corresponding current polymerization reaction type, current reaction temperature, current reaction pressure, current light type, and current light intensity; and perform one-hot encoding on the current polymerization reaction type and the current light type respectively to obtain the corresponding polymerization reaction type encoding and light intensity encoding; and perform normalization processing on the current reaction temperature based on a preset reaction temperature range to obtain the corresponding reaction temperature encoding; and perform normalization processing on the current reaction pressure based on a preset reaction pressure range to obtain the corresponding reaction pressure encoding; and perform normalization processing on the current light intensity based on a preset light intensity range to obtain the corresponding light intensity encoding; and form the corresponding first condition vector from the obtained polymerization reaction type encoding, reaction temperature encoding, reaction pressure encoding, light type encoding, and light intensity encoding; and input the first molecular graph sequence, the first ratio vector, and the first condition vector obtained this time into the property prediction model for corresponding N-class physical property prediction and synthesizability prediction processing to obtain the corresponding N physical property prediction quantities and one synthesizability prediction quantity; and form a second prediction vector with a length of N + 1 from the N physical property prediction quantities and the synthesizability prediction quantity obtained this time; and form a second label vector with a length of N + 1 from the N first physical property scalars and the first synthesizability score of the current evaluation record; and form a corresponding first prediction-label pair from the second prediction vector and the second label vector obtained this time; Step 88, identify whether the current evaluation record is the last first polymer record in the first evaluation set; if so, go to Step 89; if not, extract the next first polymer record in the first evaluation set as the new current evaluation record and return to Step 87; Step 89: Input all the obtained first prediction-label pairs into a preset first model evaluation function to calculate the corresponding first evaluation value; and identify whether the first evaluation value meets a preset first evaluation value range. If it does not meet, use the first polymer record of the first training set as the new current training record and return to Step 82. If it meets, confirm that the current model fine-tuning is completed. Among them, the first model evaluation function at least includes the RMSE function.
