Preparation method of polyamide acid imidization additive based on descriptor calculation and generation model
By screening additives based on descriptor-based computation and generative models, the energy consumption and performance damage problems of high-temperature processes for polyimide materials were solved, and efficient catalysis and performance optimization under low-temperature processes were achieved, thus expanding its application range.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-13
- Publication Date
- 2026-04-14
AI Technical Summary
The imidization process of existing polyimide materials requires high temperatures, resulting in high energy consumption, thermal damage to the substrate and performance degradation. Traditional additives have low screening efficiency, making it difficult to meet the requirements of low-temperature processes such as flexible electronics and affecting material performance.
By employing a descriptor-based computation and generative model approach, optimized additives are screened through optical chemical structure identification, machine learning prediction models, and generative adversarial networks. Combined with reaction pathway search and transition state calculation, highly efficient additives for catalytic cyclization are prepared.
Significantly reducing the imidization temperature minimizes the impact on the performance of polyimide films, expands their application range to heat-sensitive substrates and low-temperature processes, and improves the stability and applicability of material performance.
Smart Images
Figure CN121862249A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of polymer material design technology, specifically to a method for preparing polyamic acid imidized additives based on descriptor calculation and generation models. Background Technology
[0002] Polyimides, as a class of polymer materials with excellent thermal stability, electrical insulation, and mechanical properties, occupy an irreplaceable position in high-end fields such as flexible electronic devices, aerospace equipment, and microelectronic packaging. The core of its traditional preparation process lies in first polymerizing dianhydrides and diamine monomers to form polyamic acid (PAA), and then converting it into a stable imide structure through an imidization reaction. However, existing imidization processes have significant technical bottlenecks, severely limiting the expansion of application scenarios and performance improvement of polyimide materials.
[0003] In thermal imidization processes, temperatures typically need to be raised to above 300°C to achieve a full cyclization reaction. This high temperature not only leads to high energy consumption and long production cycles, but more importantly, it causes irreversible thermal damage to heat-sensitive substrates such as PET and PI / ITO. It may also cause excessive cross-linking of polymer segments, resulting in increased film brittleness and decreased mechanical flexibility, making it difficult to meet the requirements of low-temperature processes such as flexible electronics.
[0004] While chemical imidization effectively lowers reaction temperatures and has become an important route for the low-temperature preparation of polyimides, current technology has several limitations. On the one hand, limited by experimental costs, timelines, and traditional R&D models, the types of additives that can be explored are extremely limited, making it difficult to traverse a broad chemical space to discover candidates with superior performance. On the other hand, existing chemical imidization additives generally suffer from residue problems, significantly reducing the dielectric properties, mechanical strength, and film density of polyimide films, especially strong bases or phosphorus-based additives, which have a more pronounced negative impact on the overall material performance. Furthermore, current additive screening mainly relies on the experience of researchers and repeated trial and error, lacking systematic and efficient design and screening methods, resulting in low R&D efficiency and making it difficult to achieve synergistic optimization of imidization temperature reduction and material performance preservation. Summary of the Invention
[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: This application provides a method for preparing polyamic acid imidization additives based on descriptor calculation and generation models, including: Known polyamic acid imidizing additives are obtained, and the SMILES of the structural additives in the polyamic acid imidizing additives are identified by an optical chemical structure recognition module. After identification, the molecular descriptor of the polyamic acid imidized additive is calculated; The calculated molecular descriptors and known performance data of polyamic acid imidized additives are input into a pre-built machine learning prediction model for training, thereby establishing a quantitative structure-activity relationship between molecular structure and additive performance. Based on the trained prediction model, a set of candidate molecules is generated by combining it with a generative adversarial network. The candidate molecule set is subjected to reaction pathway search and transition state calculation to evaluate its catalytic cyclization ability and reaction selectivity. The candidate molecule set is screened through a multi-level screening mechanism, and the selected molecules are experimentally prepared and their performance characterized. The experimental data is then fed back to the machine learning model to achieve closed-loop optimization.
[0006] In one possible implementation, obtaining a known polyamic acid imidized additive and identifying the SMILES of the structural additive in the polyamic acid imidized additive using an optical chemical structure recognition module includes: Obtain known polyamic acid imidizing additives, including structural additives and SMILES additives; The structural additive in the polyamic acid imidization additive is input into the optical chemical structure recognition module for recognition, and the structural formula is converted into SMILES.
[0007] In one possible implementation, after identification is completed, the molecular descriptor of the polyamic acid imidized additive is calculated, including: After validating and standardizing the identified SMILES, the topological, electrical, and geometric descriptors of the polyamic acid imidized additive are calculated using the following formulas: in, For the topological descriptor Wiener index, These are the atom numbers in the molecular diagram. It is an atom and atoms The number of keys on the shortest path between them. For electrical descriptor dipole moment It is the first The charge of each atom, It is the first The mass of an atom It is the first The position vector of each atom relative to the center of molecular mass The radius of gyration is the geometric descriptor.
[0008] In one possible implementation, the step of inputting the calculated molecular descriptor and known performance data of polyamic acid imidized additives into a pre-built machine learning prediction model for training, and establishing a quantitative structure-activity relationship between molecular structure and additive performance, includes: The calculated molecular descriptors are used as input features, and the key performance data of the additive's imine catalytic efficiency, reaction selectivity, and thermal stability are used as output labels, which are then input into the machine learning prediction model for training. During training, the hyperparameters in the model are continuously adjusted using a multi-objective prediction loss function and an optimizer until the difference between the model prediction and the true value is minimized.
[0009] In one possible implementation, the multi-objective prediction loss function is calculated as follows: in, , , Weights for each property, Mean square error, For the predicted activation energy, is the glass transition temperature, and DOI is the degree of imidization.
[0010] In one possible implementation, the generation of a candidate molecule set based on the trained prediction model, combined with a generative adversarial network, includes: Set target performance parameters, including imideation catalytic cyclization efficiency, reaction selectivity threshold, and thermal stability range; The target performance parameters are input into the trained prediction model, and the core molecular descriptor interval corresponding to the target performance is obtained through reverse mapping. A molecular generation model is constructed based on generative adversarial networks, and the molecular generation model includes a generator and a discriminator. Through adversarial training between the generator and the discriminator, the structural rationality and performance compliance rate of the generated molecules are optimized, and a set of candidate molecules that meet the target performance constraints is output.
[0011] In one possible implementation, adversarial training between the generator and the discriminator optimizes the structural rationality and performance compliance rate of the generated molecules, outputting a set of candidate molecules that meet the target performance constraints, including: The random noise vector and the core molecule descriptor interval are input into the generator to generate the SMILES expression for the candidate molecule; The discriminator is used to determine the structural validity of the generated SMILES expressions and remove molecules with invalid structures. At the same time, the predicted performance values of candidate molecules output by the prediction model are combined to comprehensively judge whether they meet the structural rules and target performance requirements; After the judgment is passed, the set of candidate molecules that meet the target performance constraints is output.
[0012] In one possible implementation, the candidate molecule set is subjected to reaction pathway search and transition state calculation to evaluate its catalytic cyclization ability and reaction selectivity, including: The candidate molecule set was subjected to reaction pathway search and transition state calculation on representative PAA fragments; Use automated reaction path search to generate possible reaction paths; The transition state energy, reaction energy change, and reaction rate constant during the reaction process were calculated using the DFT method. The catalytic cyclization ability and reaction selectivity of each candidate molecule were evaluated based on the calculation results.
[0013] In one possible implementation, the formulas for calculating the transition state energy, reaction energy change, and reaction rate constant in the reaction process using the DFT method are as follows: in, For the reaction energy to change, For product energy, For reactant energy, The transition state energy, For the predicted activation energy, The reaction rate constant is... Boltzmann's constant, is Planck's constant. For Gibbs free energy activation energy, The gas constant is... For temperature.
[0014] In one possible implementation, the candidate molecule set is screened through a multi-level screening mechanism, and the preferred molecules are experimentally prepared and their performance characterized. The experimental data is then fed back to a machine learning model to achieve closed-loop optimization, including: Set preset scoring criteria; The candidate molecule set is comprehensively scored according to the preset scoring criteria; The preset number of candidate molecules with the highest scores under the preset scoring criteria are selected as the preferred molecules; The preferred molecules were added to a PAA solution at predetermined amounts to prepare thin films and their performance was characterized. Experimental data is fed back into the machine learning model to achieve closed-loop optimization.
[0015] Compared with the prior art, the beneficial effects of this application are as follows: This application establishes a quantitative structure-activity relationship between molecular structure and additive properties. Combined with reaction pathway searching and transition state calculations, the prepared additive can significantly reduce the imidization temperature of polyamic acid while minimizing negative impacts on the mechanical properties, thermal stability, dielectric properties, and film density of polyimide films. This synergistic optimization effect makes polyimide materials suitable for scenarios such as the processing of thermosensitive substrates and low-temperature production processes, greatly expanding its application scope in flexible electronics, microelectronic packaging, and other fields.
[0016] This application utilizes DFT calculations to precisely evaluate the catalytic cyclization ability and reaction selectivity of candidate molecules, effectively avoiding side reactions and ensuring the efficient and targeted imidization reaction. Simultaneously, through a multi-level screening mechanism and experimental verification, the performance reliability and stability of the selected preferred molecules are significantly improved, providing strong support for the large-scale production and high-quality application of polyimide materials.
[0017] This application establishes a closed-loop R&D system of design-prediction-experiment-optimization by feeding experimental preparation and performance characterization data into a machine learning model. This system can continuously optimize model parameters and molecular design logic, thereby continuously improving the accuracy and performance superiority of additive design, and providing a reusable and scalable technical framework for the subsequent development of novel high-performance polyamic acid imidized additives.
[0018] The additives prepared in this application are not only suitable for traditional chemical imidization systems, but also adaptable to the modification needs of photothermal triggered imidization systems and thermal generator systems, demonstrating broad applicability. Their excellent catalytic performance and material compatibility provide a universal solution for improving the performance of polyimide materials in different process scenarios, and have significant industrial application value. Attached Figure Description
[0019] Figure 1 A schematic flowchart illustrating a method for preparing polyamic acid imidization additives based on descriptor calculation and generation models, provided for embodiments of this application; Figure 2 A schematic diagram of a molecular structure recognizer provided in an embodiment of this application; Figure 3 This is a schematic diagram of a gridded molecular two-dimensional structure provided in an embodiment of this application; Figure 4 The descriptor comparison histogram provided in the embodiments of this application. Detailed Implementation
[0020] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.
[0021] Figure 1 A schematic flowchart illustrating a method for preparing polyamic acid imidization additives based on descriptor calculation and generation models, provided in this application embodiment, is shown below. Figure 1 This embodiment describes a method for preparing polyamic acid imidization additives based on descriptor computation and generation models, comprising: S101, Obtain known polyamic acid imidized additives, and identify the SMILES of the structural additives in the polyamic acid imidized additives through an optical chemical structure recognition module.
[0022] Known imidization promoters, catalysts, and chemical imidizing agents, such as imidazoles, phosphoramides, guanidines, trialkylamines, PBG / TBG systems, and related experimental data, such as the decrease in imidization temperature and changes in mechanical or thermal properties, were collected. In this embodiment, the polyamic acid imidizing additive data obtained included quinoline, triethylamine, isoquinoline, 1,4-diazabicyclo(2.2.3)octane, 5,6,7,8-tetrahydroquinoline, etc. First, the known polyamic acid imidizing additives with unknown smiles were converted from their structural formulas to smiles using an optical chemical structure recognition module, such as... Figure 2 As shown.
[0023] S102, After identification is completed, the molecular descriptor of the polyamic acid imidized additive is calculated.
[0024] In this embodiment, the identified SMILES are validated and standardized to remove invalid structures and duplicate data, thus constructing a standardized additive dataset. Quantum chemical calculation methods are used to calculate the molecular descriptors of the polyamic acid imidized additives. These molecular descriptors include topological descriptors, electronic descriptors, and geometric descriptors, calculated using the following formulas: in, For the topological descriptor Wiener index, These are the atom numbers in the molecular diagram. It is an atom and atoms The number of keys on the shortest path between them. For electrical descriptor dipole moment It is the first The charge of each atom, It is the first The mass of an atom It is the first The position vector of each atom relative to the center of molecular mass This is the radius of gyration for the geometry descriptor. See also: Figure 3 combine Figure 4 It displays the specific gridded molecular two-dimensional structure and descriptive comparison histograms and other information through a visual interface.
[0025] In addition, the formula for calculating molecular descriptors also includes: in, Molecular weight For atoms atomic mass, The topological polar surface area. This refers to the number of times a specific type of polar segment, such as hydroxyl or amino groups, appears in a molecule. This is the empirical contribution of this segment to the polar surface area. The net charge of atom A. The nuclear charge number of atom A. These are the diagonal elements of the density matrix. The off-diagonal elements of the density matrix, These are overlapping matrix elements. These descriptors can be used to predict catalytic cyclization tendency (reaction coordinate energy barrier) and material compatibility indices (possible side reactions with polyimide groups, residual thermal decomposition temperature).
[0026] S103, the calculated molecular descriptor and known performance data of polyamic acid imidized additives are input into a pre-built machine learning prediction model for training, and a quantitative structure-activity relationship between molecular structure and additive performance is established.
[0027] In this embodiment, the calculated molecular descriptor is used as the input feature, and the key performance data of the additive's imine catalytic efficiency, reaction selectivity, and thermal stability are used as the output label. These are input into the machine learning prediction model for training. The prediction model employs a hybrid model architecture of XGBoost and neural networks. XGBoost is used to capture the linear correlation and local feature interactions between the molecular descriptor and performance, while the neural network is used to uncover nonlinear complex mapping relationships. During training, the hyperparameters in the model are continuously adjusted using a multi-objective prediction loss function and an optimizer until the difference between the model prediction and the true value is minimized. The formula for calculating the multi-objective prediction loss function is as follows: in, , , Weights for each property, Mean square error, For the predicted activation energy, Here, represents the glass transition temperature, and DOI represents the degree of imidization. The core function of the predictive model is to establish a quantitative structure-activity relationship between molecular structure and additive properties, thus realizing the fundamental function of predicting performance from structure.
[0028] In this embodiment, a random forest prediction model can also be used for prediction. The random forest prediction model is as follows: in, This represents the final predicted value of the model, where N is the total number of decision trees, and the symbol is... Representing the A decision tree, Represents a vector of molecular descriptors.
[0029] S104 generates a set of candidate molecules based on the trained prediction model and combined with a generative adversarial network.
[0030] In this embodiment, target performance parameters are set, including imine catalytic cyclization efficiency, reaction selectivity threshold, and thermal stability range. These target performance parameters are input into a trained prediction model, and a core molecular descriptor interval corresponding to the target performance is obtained through back-mapping. A molecular generation model is constructed based on a generative adversarial network, which includes a generator and a discriminator. Random noise vectors and core molecular descriptor intervals are input into the generator to generate SMILES expressions for candidate molecules. The discriminator performs structural validity judgment on the generated SMILES expressions, eliminating structurally invalid molecules. Simultaneously, the predicted performance values of candidate molecules output by the prediction model are combined to comprehensively determine whether they meet the structural rules and target performance requirements. Through adversarial training between the generator and the discriminator, the structural rationality and performance compliance rate of the generated molecules are optimized, and a set of candidate molecules that meet the target performance constraints is output.
[0031] In this embodiment, deep generation algorithms such as Conditional Variational Autoencoder (CVAE), molecular diffusion models, and SMILES-RNN can also be used to generate a large number of candidate molecules. The generation method involves an agent that perceives the topological state and charge distribution of the current molecular structure in real time, and presets conditional variables such as the desired range of the target activation energy. The Actor network selects the next step with the highest probability from the action space based on the current molecular state, attaching specific chemical functional groups to adjust the imidization promoting ability. The PPO algorithm uses shear probability ratios to ensure that the change in molecular structure is not too aggressive, maintaining the stability of the generation process. Then, a multi-objective reward mechanism is used to comprehensively score the generated molecules, and Pareto front analysis is used to find the optimal balance between imidization promoting ability and material compatibility.
[0032] The formula for the VAE optimization objective is: In the formula, Let be the objective function. For expectation operator, To reconstruct the likelihood, x is the molecular sequence diagram. z is the latent space vector. For encoder distribution, For decoder distribution, Let KL divergence be denoted as KL divergence.
[0033] The formula for the diffusion model is: In the formula, For conditional probability distribution, It follows a multivariate normal distribution. The mean of a normal distribution is . Encoding the original molecule, The variance of the normal distribution is... This is the encoding after adding noise. This is the noise accumulation factor.
[0034] S105 performs reaction pathway search and transition state calculation on the candidate molecule set to evaluate the catalytic cyclization ability and reaction selectivity of the candidate molecule set.
[0035] In this embodiment, reaction pathway search and transition state calculation are performed on the candidate molecule set on a representative PAA fragment. Automated reaction pathway search methods such as AFIR and AutoMeKin are used to generate possible reaction pathways. The DFT method is used to calculate the transition state energy, reaction energy change, and reaction rate constant during the reaction process. The calculation formulas are as follows: in, For the reaction energy to change, For product energy, For reactant energy, The transition state energy, For the predicted activation energy, The reaction rate constant is... Boltzmann's constant, is Planck's constant. For Gibbs free energy activation energy, The gas constant is... The temperature is used as the criterion. The catalytic cyclization ability and reaction selectivity of each candidate molecule are evaluated based on the calculation results. In this embodiment, a reaction relationship network is constructed and the difference in competing paths is calculated to evaluate the reaction selectivity. Alternatively, the catalytic cyclization ability of each candidate molecule can be evaluated using the reduction in reaction energy barrier as the core indicator, and the reaction selectivity can be evaluated using the proportion of the target cyclized product as the indicator, while eliminating candidate molecules with excessively high reaction energy barriers and unsatisfactory selectivity.
[0036] S106 uses a multi-level screening mechanism to screen the candidate molecule set and conduct experimental preparation and performance characterization of the selected molecules, and feeds the experimental data back to the machine learning model to achieve closed-loop optimization.
[0037] In this embodiment, the generated polyimide additive candidate molecules are comprehensively scored. The scoring criteria are: activation energy reduction ratio greater than or equal to 20%, syntheticity score greater than or equal to 0.7, predicted mechanical property retention rate greater than or equal to 90%, and thermal stability score greater than or equal to 85%. Under these criteria, the top five candidate molecules with the highest scores are added to PAA solution at 0.1–5 wt%, using NMP or DMAc as solvents, to prepare films and perform the following tests: DSC to determine the imidization initiation and peak temperatures; FTIR to detect the amide → imide conversion rate; tensile testing to evaluate mechanical properties; and TGA to evaluate thermal stability. Feedback and closed-loop optimization are then performed, and the experimental results are fed back to the machine learning model to update the weights, forming a closed-loop optimization system.
[0038] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0039] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0040] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.
Claims
1. A method for preparing polyamic acid imidization additives based on descriptor computation and generation models, characterized in that, include: Known polyamic acid imidizing additives are obtained, and the SMILES of the structural additives in the polyamic acid imidizing additives are identified by an optical chemical structure recognition module. After identification, the molecular descriptor of the polyamic acid imidized additive is calculated; The calculated molecular descriptors and known performance data of polyamic acid imidized additives are input into a pre-built machine learning prediction model for training, thereby establishing a quantitative structure-activity relationship between molecular structure and additive performance. Based on the trained prediction model, a set of candidate molecules is generated by combining it with a generative adversarial network. The candidate molecule set is subjected to reaction pathway search and transition state calculation to evaluate its catalytic cyclization ability and reaction selectivity. The candidate molecule set is screened through a multi-level screening mechanism, and the selected molecules are experimentally prepared and their performance characterized. The experimental data is then fed back to the machine learning model to achieve closed-loop optimization.
2. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, The process of obtaining known polyamic acid imidized additives, and identifying the SMILES of structural additives in the polyamic acid imidized additives using an optical chemical structure recognition module, includes: Obtain known polyamic acid imidizing additives, including structural additives and SMILES additives; The structural additive in the polyamic acid imidization additive is input into the optical chemical structure recognition module for recognition, and the structural formula is converted into SMILES.
3. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, After identification, the molecular descriptor of the polyamic acid imidized additive is calculated, including: After validating and standardizing the identified SMILES, the topological, electrical, and geometric descriptors of the polyamic acid imidized additive are calculated using the following formulas: in, For the topological descriptor Wiener index, These are the atom numbers in the molecular diagram. It is an atom and atoms The number of keys on the shortest path between them. For electrical descriptor dipole moment It is the first The charge of each atom, It is the first The mass of an atom It is the first The position vector of each atom relative to the center of molecular mass The radius of gyration is the geometric descriptor.
4. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, The step of inputting the calculated molecular descriptor and known performance data of polyamic acid imidized additives into a pre-built machine learning prediction model for training, and establishing a quantitative structure-activity relationship between molecular structure and additive performance, includes: The calculated molecular descriptors are used as input features, and the key performance data of the additive's imine catalytic efficiency, reaction selectivity, and thermal stability are used as output labels, which are then input into the machine learning prediction model for training. During training, the hyperparameters in the model are continuously adjusted using a multi-objective prediction loss function and an optimizer until the difference between the model prediction and the true value is minimized.
5. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 4, characterized in that, The formula for calculating the multi-objective prediction loss function is: in, , , Weights for each property, Mean square error, For the predicted activation energy, is the glass transition temperature, and DOI is the degree of imidization.
6. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, The prediction model, based on training, is combined with a generative adversarial network to generate a set of candidate molecules, including: Set target performance parameters, including imideation catalytic cyclization efficiency, reaction selectivity threshold, and thermal stability range; The target performance parameters are input into the trained prediction model, and the core molecular descriptor interval corresponding to the target performance is obtained through reverse mapping. A molecular generation model is constructed based on generative adversarial networks, and the molecular generation model includes a generator and a discriminator. Through adversarial training between the generator and the discriminator, the structural rationality and performance compliance rate of the generated molecules are optimized, and a set of candidate molecules that meet the target performance constraints is output.
7. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 6, characterized in that, Through adversarial training between the generator and the discriminator, the structural rationality and performance compliance rate of the generated molecules are optimized, and a set of candidate molecules that meet the target performance constraints is output, including: The random noise vector and the core molecule descriptor interval are input into the generator to generate the SMILES expression for the candidate molecule; The discriminator is used to determine the structural validity of the generated SMILES expressions and remove molecules with invalid structures. At the same time, the predicted performance values of candidate molecules output by the prediction model are combined to comprehensively judge whether they meet the structural rules and target performance requirements; After the judgment is passed, the set of candidate molecules that meet the target performance constraints is output.
8. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, The candidate molecule set is subjected to reaction pathway search and transition state calculation to evaluate its catalytic cyclization ability and reaction selectivity, including: The candidate molecule set was subjected to reaction pathway search and transition state calculation on representative PAA fragments; Use automated reaction path search to generate possible reaction paths; The transition state energy, reaction energy change, and reaction rate constant during the reaction process were calculated using the DFT method. The catalytic cyclization ability and reaction selectivity of each candidate molecule were evaluated based on the calculation results.
9. The method for preparing polyamic acid imidization additives based on descriptor computation and generation models according to claim 8, characterized in that, The formulas for calculating the transition state energy, reaction energy change, and reaction rate constant in the reaction process using the DFT method are as follows: in, For the reaction energy to change, For product energy, For reactant energy, The transition state energy, For the predicted activation energy, The reaction rate constant is... Boltzmann's constant, is Planck's constant. For Gibbs free energy activation energy, The gas constant is... For temperature.
10. The method for preparing polyamic acid imidization additives based on descriptor calculation and generation models according to claim 1, characterized in that, The candidate molecule set is screened through a multi-level screening mechanism, and the selected molecules are experimentally prepared and their performance characterized. The experimental data is then fed back to a machine learning model to achieve closed-loop optimization, including: Set preset scoring criteria; The candidate molecule set is comprehensively scored according to the preset scoring criteria; The preset number of candidate molecules with the highest scores under the preset scoring criteria are selected as the preferred molecules; The preferred molecules were added to a PAA solution at predetermined amounts to prepare thin films and perform performance characterization. Experimental data is fed back into the machine learning model to achieve closed-loop optimization.