Organic liquid hydrogen storage molecule design method and device

Abstract

The invention discloses an organic liquid hydrogen storage molecule design method and device. The method comprises the following steps: acquiring a first data set containing organic liquid molecule physicochemical characteristic information; constructing a splicing code based on a molecular structure in the first data set to obtain a second data set; the splicing code comprises a dehydrogenation molecule structure and a hydrogenation molecule structure; constructing a regression prediction model, and training by using the second data set; constructing a molecular design model, splicing the code input, and generating and screening new hydrogen storage molecules; inputting the new hydrogen storage molecules into the trained regression prediction model, and predicting the addition / dehydrogenation capability of the new hydrogen storage molecules to obtain a prediction result; and screening the molecule with the best hydrogen storage capacity based on the prediction result. By using GNN model regression prediction and VAE model molecular design, a new method is provided for novel organic liquid hydrogen storage molecular design.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material calculation, combining computational chemistry and machine learning, and particularly relates to a technical scheme for jointly using a graph neural network and a variational autoencoder for design and screening of organic liquid hydrogen storage molecules. BACKGROUND

[0002] Hydrogen energy, as a clean and high-energy-density secondary energy, is widely recognized as an important carrier for realizing low-carbon energy system transformation in the future. How to efficiently, safely and reversibly store and release hydrogen has always been a key problem in the hydrogen energy technology chain. Among the many hydrogen storage schemes, the organic liquid hydrogen carrier (LOHC) system has attracted widespread attention due to its liquid operation, reusability, compatibility with existing petrochemical infrastructure and other advantages. LOHC molecules realize the hydrogen absorption and release process through reversible hydrogenation / dehydrogenation reactions, have a high volumetric hydrogen storage density, and can achieve safe transportation under normal pressure, and are therefore considered as a highly industrial potential hydrogen storage technology route.

[0003] However, the design of LOHC molecules presents significant challenges. Traditional hydrogen storage molecule screening often relies on density functional theory (DFT) calculations or experimental measurements. The former has high accuracy but high computational cost and low throughput, while the latter requires complex synthesis and characterization, has a long cycle and consumes a lot of resources. At the same time, the performance of molecular hydrogen storage is affected by multiple factors, such as: the electronic structure and aromatic stability of the molecule; the hydrogenation / dehydrogenation reaction path and transition state energy barrier; the influence of hydrogenation degree on thermodynamic and kinetic reversibility; the balance between molecular mass and hydrogen content; In recent years, machine learning and deep learning technologies have been introduced into the field of chemistry and materials science, providing a new path for high-throughput screening and property prediction. In particular, graph neural networks (GNN) can directly learn on molecular graph structures, treating atoms as nodes and chemical bonds as edges, and automatically capturing local chemical environments and global structural features. Compared with traditional molecular descriptor methods (such as molecular fingerprints, Coulomb matrices, etc.), GNN can learn the structure-property mapping relationship in an end-to-end manner, significantly improving the prediction accuracy and interpretability of the model. In the invention patent (CN119976733A), it is pointed out that developing new liquid organic hydrogen storage materials is a pressing technical problem. To this end, the variational autoencoder (VAE) that has emerged in recent years provides a new approach to molecular generation. VAE can map complex discrete molecular structures to continuous latent space by learning the latent space distribution of molecular structures. In the latent space, smooth interpolation, random sampling or directed optimization can be realized to generate new molecules with target properties.

[0004] However, relative to the design of molecules, there are few existing models for the design of organic liquid hydrogen storage molecules, and there is currently no machine learning model for the design of hydrogen storage molecules. In this context, the present application proposes a GNN model combined with the generation of molecular information, the regression prediction function and the molecular design function of the VAE model, for the precise design of organic liquid hydrogen storage molecules. SUMMARY

[0005] In view of the deficiencies of existing research, the present application provides a calculation method and system based on the combination of GNN and VAE, for the high-throughput design of organic liquid hydrogen storage molecules.

[0006] In a first aspect, an organic liquid hydrogen storage molecule design method is provided, comprising: obtaining a first data set containing the physicochemical property information of the organic liquid molecule; constructing a splicing code based on the molecular structure in the first data set to obtain a second data set; the splicing code includes dehydrogenated molecules and hydrogenated molecule structures; constructing a regression prediction model and training it using the second data set; constructing a molecular design model to input the splicing code, generate and screen new hydrogen storage molecules; inputting the new hydrogen storage molecules into the trained regression prediction model to predict the hydrogenation / dehydrogenation capacity of the new hydrogen storage molecules, and obtaining the prediction results; screening the molecules with the best hydrogen storage capacity based on the prediction results.

[0007] In a possible implementation manner, the physicochemical properties include atomic structure, atomic number, atomic charge, electronegativity, covalent radius, hybrid state, aromaticity and atomic measure.

[0008] In a possible implementation manner, the method for constructing the splicing code comprises: obtaining a dehydrogenated molecule structure G d and a hydrogenated molecule structure G h based on the original molecular structure, respectively; and splicing the dehydrogenated molecule structure G d and the hydrogenated molecule structure G to obtain the splicing code.

[0009] In a possible implementation manner, the regression prediction model includes a graph neural network (GNN), and the training method comprises: taking the splicing code as the input and the key features affecting the hydrogenation / dehydrogenation reaction performance as the output, and training using the second data set.

[0010] In a possible implementation manner, the molecular design model includes a variational autoencoder (VAE).

[0011] In a possible implementation manner, the molecular design model is constructed to input the spliced encoding, generate and screen a new hydrogen storage molecule, and the method comprises the following steps of: inputting the spliced encoding into an encoder of a variational autoencoder VAE to obtain a latent representation space z of a reaction molecular structure; wherein the latent representation space z adds noise d; the latent representation space z is decoded by a decoder to obtain a SMILES molecular graph; confirming rationality of the molecule, and screening the new hydrogen storage molecule.

[0012] In a second aspect, an organic liquid hydrogen storage molecule design device is provided, and the device comprises: an acquisition module configured to acquire a first data set containing physical and chemical property information of an organic liquid molecule; a splicing module configured to construct spliced encoding based on a molecular structure in the first data set to obtain a second data set; the spliced encoding comprises a dehydrogenation molecule and a hydrogenation molecule structure; a first construction module configured to construct a regression prediction model and train the regression prediction model by using the second data set; a second construction module configured to construct a molecular design model to input the spliced encoding, generate and screen a new hydrogen storage molecule; a prediction module configured to input the new hydrogen storage molecule into the trained regression prediction model to predict hydrogenation / dehydrogenation capacity of the new hydrogen storage molecule, and obtain a prediction result; an output module configured to screen a molecule with the best hydrogen storage capacity based on the prediction result.

[0013] In a third aspect, an electronic device is provided, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the organic liquid hydrogen storage molecule design method according to the first aspect when executing the program.

[0014] In a fourth aspect, a non-transitory computer readable storage medium is provided, which stores a computer program, and the computer program is executable by a processor to implement the organic liquid hydrogen storage molecule design method according to the first aspect.

[0015] In a fifth aspect, a computer program product is provided, which comprises a computer program, and the computer program is executable by a processor to implement the organic liquid hydrogen storage molecule design method according to the first aspect.

[0016] The beneficial effects of the present application are as follows: 1) The method provided by the present application realizes high-throughput prediction under the premise of ensuring accuracy; 2) The method provided by the present application can upgrade from “passive prediction” to “active design” through the combination of GNN and VAE models, and greatly expand the selection range of candidate organic molecules with high hydrogen storage capacity. 3) The method provided by the present application significantly improves the calculation efficiency and reduces the operation cost. Compared with directly using DFT to perform full space search, GNN-VAE only needs a small amount of data as initial training samples to predict new molecular structures. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Figure 1 The flowchart of the method provided by the embodiment of the present application is shown in the figure. Figure 2 The schematic diagram of the GNN regression prediction model described in the embodiment of the present application is shown in the figure. Figure 3 The schematic diagram of the VAE molecular design model architecture described in the embodiment of the present application is shown in the figure. Figure 4 The schematic diagram of the organic liquid hydrogen storage molecular design device provided by the embodiment of the present application is shown in the figure. Figure 5 The schematic diagram of the electronic device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present application more clear, the technical solutions in the present application will be described clearly and completely in the following with reference to the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] At present, there are few existing models for designing organic liquid hydrogen storage molecules, and there is no method that can simultaneously design new hydrogen storage molecules and predict hydrogen storage performance.

[0021] In view of this, the present application provides an organic liquid hydrogen storage molecular design method and device for accurately designing and predicting the hydrogen storage performance of organic liquid hydrogen storage molecules.

[0022] Referring to Figure 1 The present application provides an organic liquid hydrogen storage molecular design method based on machine learning, which comprises: S101, obtaining a first data set containing the physicochemical property information of organic liquid molecules.

[0023] In a possible implementation manner, the physical and chemical properties include molecular structure, atomic number, atomic charge, electronegativity, covalent radius, hybrid state, aromaticity, and atomic measure.

[0024] S102, construct a splicing code based on the molecular structure in the first data set to obtain a second data set; the splicing code includes a dehydrogenation molecule and a hydrogenation molecule structure.

[0025] In a possible implementation manner, the method for constructing the splicing code includes: obtaining a dehydrogenation molecule structure G d and a hydrogenation molecule structure G h based on the original molecule structure respectively. d The dehydrogenation molecule structure G pair and the hydrogenation molecule structure G are spliced to obtain the splicing code.

[0026] Further, the format of the splicing code z pair is as follows: z d = [G h ].

[0027] It can be understood that, compared with single-atom encoding, the splicing code has the following advantages: 1. Traditional single-molecule encoding can only output single properties, and the reaction energy needs to be obtained by subtracting the predictions of two models; and the splicing code model directly learns the energy difference function between two molecules, avoids error accumulation, and the model can more directly capture the influence of structural changes on the energy difference; 2. The splicing input allows the GNN to learn the change pattern of corresponding atoms / bonds; the learning process is more in line with the chemical reactant-product pairing logic; and it is helpful for the model to identify the reaction center and the reaction path; 3. The splicing input provides a more consistent and interpretable latent space representation for subsequent generation models (VAE decoder).

[0028] S103, construct a regression prediction model and train the regression prediction model by using the second data set.

[0029] In a possible implementation manner, the regression prediction model includes a graph neural network GNN.

[0030] In a possible implementation manner, the training method includes: taking the splicing code z pair as input, taking the key features affecting the performance of the dehydrogenation / hydrogenation reaction as output, and training the regression prediction model by using the second data set.

[0031] Further, the key features affecting the performance of the dehydrogenation / hydrogenation reaction include Gibbs free energy ΔG, aromaticity, zero-point vibration energy, physical property, ground state internal energy, enthalpy, and polarizability.

[0032] It can be understood that the key features are selected as one output or weighted as an output. The key features are selected and weighted according to the needs of the target molecule.

[0033] S104, constructing a molecular design model to splice the encoding input to generate and screen new hydrogen storage molecules.

[0034] In one possible implementation manner, S104 includes: S104a, inputting the spliced encoding into the encoder of the variational autoencoder VAE to obtain a latent representation space z of the reaction molecular structure; wherein the latent representation space z adds noise d; S104b, the latent representation space z can obtain a SMILES molecular graph after decoding by the decoder; S104c, confirming the rationality of the molecule to screen new hydrogen storage molecules.

[0035] S105, inputting the new hydrogen storage molecule into the trained regression prediction model to predict the hydrogen storage capacity of the new hydrogen storage molecule, and obtaining a prediction result.

[0036] S106, screening a molecule with the best hydrogen storage capacity based on the prediction result.

[0037] The following will be described in more detail in conjunction with embodiments.

[0038] An organic liquid hydrogen storage molecule design method based on machine learning, comprising: S1, obtaining a first data set containing the physicochemical property information of the organic liquid molecule.

[0039] The physicochemical property information of the organic liquid molecule in the embodiment of the application is from the QM9 data set, and the basic information of the related organic liquid hydrogen storage molecule form is extracted in QM9 by searching. The physicochemical properties include atomic structure, atomic number, atomic charge, electronegativity, covalent radius, hybrid state, aromaticity and atomic measure.

[0040] S2, constructing a splicing code based on the molecular structure in the first data set to obtain a second data set; the splicing code includes dehydrogenated molecule and hydrogenated molecule structure.

[0041] Specifically, based on the original molecular structure, the dehydrogenated molecule SMILE format (Gd) and the hydrogenated molecule (Gh) are obtained respectively; the dehydrogenated molecule structure Gd and the hydrogenated molecule structure G are spliced to obtain the splicing code Z pair =[Gd||Gh].

[0042] S3, constructing a regression prediction model and training using the second data set.

[0043] Specifically, the regression prediction model adopts a GNN model.

[0044] Specifically, referring to Figure 2 , the training method is as follows: Z pair The joint features are input into the GNN prediction network, and the output is: y' = f(Z pair ), y' is the key feature in the data set that affects the performance of the hydrogenation / dehydrogenation reaction.

[0045] The mean square error MSE is used to measure the difference between the model prediction and the true value; the Adam optimizer is used to update the parameters of the GNN to minimize the loss function, thereby improving the prediction accuracy.

[0046] After training, the GNN can predict the key features that affect the performance of the hydrogenation / dehydrogenation reaction from the molecular structure, including: ΔG, aromaticity, zero-point vibration energy, physical properties, ground state internal energy, enthalpy, and polarizability.

[0047] S4, constructing a molecular design model to splice the encoded input and generate and screen new hydrogen storage molecules.

[0048] Specifically, the molecular design model adopts a VAE, and its task is to generate new hydrogen storage molecules.

[0049] Referring to Figure 3 , the specific steps are as follows: Z pair in S2 is input into the encoder of the VAE to obtain the latent representation space z of the reaction molecular structure, and the latent representation space z is added with noise d, z = σ + μδ + d; wherein, σ represents the confidence of the model to the feature, μ represents the average chemical feature of the molecule in the latent space, and δ represents random noise subject to a standard normal distribution N(0, 1).

[0050] The structure of the encoder of the VAE can ensure the continuity and samplability of the latent space, and can introduce controlled randomness while maintaining the chemical rationality of the original molecule, so as to generate new molecules with diverse structures and potential high hydrogen storage capacity in the decoding process. Adding noise d can ensure the accuracy of the latent representation space z.

[0051] After the latent representation space z is decoded by the decoder, a SMILES molecular graph can be obtained, and then the rationality of the generated molecule is judged by the RDkit tool package.

[0052] S5, input the designed new molecular code into the regression model trained in S3, which is used to predict the hydrogenation / dehydrogenation capacity of the generated new molecule.

[0053] S6, the new molecules obtained in the regression prediction model are sorted and screened, and the molecular structure with the best hydrogen storage capacity is selected.

[0054] The organic liquid hydrogen storage molecule design device provided in the present application is described below. The organic liquid hydrogen storage molecule design device described below can be referred to in conjunction with the organic liquid hydrogen storage molecule design method described above.

[0055] Figure 4 is a structural schematic diagram of the organic liquid hydrogen storage molecule design device provided in the embodiments of the present application, as Figure 4 shown, comprising an acquisition module 41, a splicing module 42, a first construction module 43, a second construction module 44, a prediction module 45 and an output module 46, wherein: The acquisition module 41 is configured to acquire a first data set containing physical and chemical property information of organic liquid molecules. The splicing module 42 is configured to construct a splicing code based on the molecular structure in the first data set to obtain a second data set. The splicing code includes dehydrogenated molecules and hydrogenated molecules. The first construction module 43 is configured to construct a regression prediction model and train the second data set. The second construction module 44 is configured to construct a molecular design model, input the splicing code, and generate and screen new hydrogen storage molecules. The prediction module 45 is configured to input the new hydrogen storage molecules into the trained regression prediction model to predict the hydrogenation / dehydrogenation capacity of the new hydrogen storage molecules and obtain a prediction result. The output module 46 is configured to screen the molecules with the best hydrogen storage capacity based on the prediction result.

[0056] Figure 5 An example of an entity structure schematic diagram of an electronic device is shown in Figure 5 The electronic device can include a processor 510, a communications interface 520, a memory 530 and a communications bus 540. The processor 510, the communications interface 520 and the memory 530 can communicate with each other through the communications bus 540. The processor 510 can invoke the logical instructions in the memory 530 to execute the organic liquid hydrogen storage molecule design method.

[0057] Further, the logic instructions in the memory 530 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0058] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, so that the computer can execute the organic liquid hydrogen storage molecule design method provided by the above-mentioned methods.

[0059] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the organic liquid hydrogen storage molecule design method provided by the above-mentioned methods.

[0060] The device embodiments described above are only schematic, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0061] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the technical solutions described above or the parts that make contributions to the prior art essentially can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0062] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for designing organic liquid hydrogen storage molecules, characterized in that, include: Obtain the first dataset containing information on the physicochemical properties of organic liquid molecules; The second dataset is obtained by constructing splicing codes based on the molecular structures in the first dataset; the splicing codes include dehydrogenated and hydrogenated molecular structures. Build a regression prediction model and train it using the second dataset; A molecular design model is constructed to generate and screen new hydrogen storage molecules by splicing coded inputs. The new hydrogen storage molecule is input into the trained regression prediction model to predict its hydrogen addition / dehydrogenation capability, and the prediction results are obtained. Based on the prediction results, the molecule with the best hydrogen storage capacity was selected.

2. The method for designing organic liquid hydrogen storage molecules according to claim 1, characterized in that, The physicochemical properties include molecular structure, atomic number, atomic charge, electronegativity, covalent radius, hybridization state, aromaticity, and atomic measurement.

3. The method for designing organic liquid hydrogen storage molecules according to claim 1, characterized in that, The method for constructing splicing codes includes: obtaining the dehydrogenated molecular structure G based on the original molecular structure. d and hydrogenation molecular structure G h ; the dehydrogenation molecular structure G d The splicing code is obtained by splicing the hydrogenated molecular structure G together.

4. The method for designing organic liquid hydrogen storage molecules according to claim 1, characterized in that, The regression prediction model includes a graph neural network (GNN); The training method includes: using concatenated encoding as input and key features affecting the performance of addition / dehydrogenation reactions as output, and training with a second dataset.

5. The method for designing organic liquid hydrogen storage molecules according to claim 1, characterized in that, The molecular design model includes a variational autoencoder (VAE).

6. The method for designing organic liquid hydrogen storage molecules according to claim 1, characterized in that, The construction of the molecular design model, using spliced ​​coded input, generates and screens new hydrogen storage molecules, including: The splicing code is input into the encoder of the variational autoencoder (VAE) to obtain the latent representation space z of the reactive molecular structure; where noise d is added to the latent representation space z. The latent representation space z can be decoded by the decoder to obtain the SMILES molecular map; The rationale for the molecule was confirmed, and new hydrogen storage molecules were screened out.

7. An organic liquid hydrogen storage molecule design device, characterized in that, include: The acquisition module is used to acquire a first dataset containing information on the physicochemical properties of organic liquid molecules. The splicing module is used to construct splicing codes based on the molecular structures in the first dataset to obtain the second dataset; the splicing codes include dehydrogenated and hydrogenated molecular structures; The first building block is used to build a regression prediction model, which is trained using the second dataset; The second building module is used to build molecular design models to splice coded inputs, generate and screen new hydrogen storage molecules; The prediction module is used to input new hydrogen storage molecules into the trained regression prediction model to predict the hydrogen addition / dehydrogenation capability of the new hydrogen storage molecules and obtain the prediction results. The output module is used to screen out molecules with the best hydrogen storage capacity based on the prediction results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the organic liquid hydrogen storage molecule design method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the organic liquid hydrogen storage molecule design method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the organic liquid hydrogen storage molecule design method as described in any one of claims 1 to 6.

Images