Redox potential prediction method based on carbonyl compound descriptor

Abstract

The invention belongs to the field of flow batteries, and particularly relates to a molecular feature descriptor capable of being used for oxidation-reduction potential of a carbonyl compound, which is mainly used for predicting the oxidation-reduction potential of the carbonyl compound. Comprising the following steps: collecting and sorting structural information of a large number of carbonyl compounds, and selecting part of molecules to obtain redox potential data of the molecules through DFT calculation; on the basis of the molecular structure data, constructing a molecular feature descriptor provided by the invention; then, the extracted descriptors are trained through a machine learning algorithm, and an efficient and accurate prediction model is established; when the model is used for predicting the additional carbonyl compound, it is proved that the oxidation-reduction potential of the carbonyl compound can be rapidly and accurately predicted, and an important reference basis is provided for design and optimization of the carbonyl compound. The descriptor provided by the invention has the advantages of large information amount, high prediction accuracy, high calculation efficiency and the like, and can provide powerful support for research and development of carbonyl compounds for flow batteries.

Description

Technical Field

[0001] The present invention belongs to the field of liquid flow batteries, and in particular is a descriptor for predicting the redox potential of carbonyl compounds. Background Art

[0002] As the demand for energy in modern society continues to grow, flow batteries have attracted more and more attention as a technology with efficient energy storage capabilities. The potential of the redox reaction involved in the electrode reaction is crucial to the energy density of the flow battery. Therefore, it is urgent to find an organic electrolyte material with high redox potential, high water solubility and fast redox kinetics. Among the reported organic electrolyte materials, organic carbonyl compounds (such as quinones) are considered to be one of the most promising organic electrolyte materials due to their high theoretical capacity and fast reaction kinetics. They have been widely studied in recent years. However, in the face of the huge number of carbonyl compounds and their complexity, it is necessary to develop reasonable design strategies and accelerate the discovery of such materials. However, it is impossible to synthesize and characterize such a large number of materials solely relying on the experience and intuition of researchers. At the same time, due to the limitations of time and computing resources, it is difficult to fully explore all possibilities and diversity using traditional quantum chemical calculation methods. Machine learning (ML)-assisted screening of large databases is a promising approach. However, for the task of redox potential prediction, the prediction accuracy of ML models still needs to be improved, and the use of more informative feature descriptors is a promising approach. Summary of the invention

[0003] The present invention provides a descriptor that can be used to predict the redox potential of carbonyl compounds. The descriptor combines the characteristics of carbonyl compounds and their corresponding alcohol compounds, and can directly predict their redox potential based on their molecular structures. The descriptor for predicting the redox potential of carbonyl compounds for flow batteries designed using the method of the present invention can significantly improve the accuracy of machine learning models for predicting the redox potential of carbonyl compounds.

[0004] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a method for predicting redox potential based on carbonyl compound descriptors, comprising the following steps:

[0005] Step 1: Screen carbonyl compound molecules based on element type, number of rings, and number of heavy atoms;

[0006] Step 2: Randomly select some of the molecules to calculate the redox potential, and construct a training data set based on the carbonyl compound molecules and their corresponding redox potentials;

[0007] Step 3: Generate an oxidized state molecular feature descriptor and a reduced state molecular feature descriptor based on the molecular descriptor, and then construct a descriptor for the carbonyl compound according to the difference between the oxidized state molecular feature descriptor and the reduced state molecular feature descriptor;

[0008] Step 4: Use the newly generated descriptors in the machine learning model and train the data in the training set using the GP regression algorithm;

[0009] Step 5: The carbonyl compound to be tested is subjected to step 3 to obtain the descriptor, and then the predicted redox potential is obtained through the trained machine learning model.

[0010] The screening includes:

[0011] (1) Contains only H, C, O, N, P, S, F, and Cl elements;

[0012] (2) The number of molecular rings is less than 3. For molecules with 0 rings, the number of heavy atoms other than hydrogen atoms is less than or equal to 6. For molecules with 1 ring, the number of heavy atoms is less than or equal to 6. For molecules with 2 rings, the number of heavy atoms is less than or equal to 16. For molecules with 3 rings, the number of heavy atoms is less than or equal to 20.

[0013] (3) The molecule contains carbonyl groups, which cannot all come from ester groups;

[0014] (4) It does not contain carboxyl or phosphonic acid groups.

[0015] The step 3: generating an oxidized state molecular feature descriptor and a reduced state molecular feature descriptor based on the molecular descriptor, and then constructing a descriptor for carbonyl compounds according to the difference between the oxidized state molecular feature descriptor and the reduced state molecular feature descriptor, as follows:

[0016] Based on Mordred descriptor, RDKit descriptor, and Morgan molecular fingerprint, the characteristic descriptor of the oxidized state molecule is subtracted from the characteristic descriptor of the reduced state molecule in the same dimension;

[0017] If the result is not 0, the subtracted feature is used as the descriptor for the carbonyl compound;

[0018] If the result is 0, the identity of the oxidized or reduced molecule is retained as the descriptor for the carbonyl compound.

[0019] A redox potential prediction system based on carbonyl compound descriptors, comprising:

[0020] The data set construction module is used to screen carbonyl compound molecules according to element type, number of rings, and number of heavy atoms; randomly select some of the molecules to calculate the redox potential, and construct a training data set based on the carbonyl compound molecules and their corresponding redox potentials;

[0021] A descriptor generation module, used for generating an oxidation state molecular feature descriptor and a reduction state molecular feature descriptor based on the molecular descriptor, and then constructing a descriptor for the carbonyl compound according to the difference between the oxidation state molecular feature descriptor and the reduction state molecular feature descriptor;

[0022] The model training module is used to use the newly generated descriptors in the machine learning model and train the data in the training set using the GP regression algorithm;

[0023] The redox potential prediction module is used to obtain the descriptor of the carbonyl compound to be tested through step 3, and then obtain the predicted redox potential through the trained machine learning model.

[0024] A redox potential prediction device based on carbonyl compound descriptors comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the redox potential prediction method based on carbonyl compound descriptors when executing the computer program.

[0025] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting redox potential based on carbonyl compound descriptors is implemented.

[0026] The present invention has the following beneficial effects and advantages:

[0027] 1. The present invention has the advantages of short time and low cost. The traditional experimental method requires a lot of synthesis and testing, which is costly and time-consuming. The method provided by the present invention can obtain the results through calculation, saving experimental cost and time.

[0028] 2. Using the descriptor of the redox potential of carbonyl compounds for liquid flow batteries described in the present invention, a large number of carbonyl compounds can be quickly and accurately predicted using a machine learning algorithm, and the predicted results have higher consistency and accuracy than experimental measurements. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of the method for constructing molecular feature descriptors proposed in the present invention;

[0030] Figure 2 Prediction accuracy result diagram of the machine learning model based on the molecular feature descriptor proposed in the present invention on the training set and the test set;

[0031] Figure 3 A graph showing the prediction accuracy of the machine learning model based on the molecular feature descriptors proposed in the present invention in two additional validation sets. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] The present invention proposes a feature descriptor combining carbonyl compounds and their corresponding alcohol compounds, and uses it to establish an ML model for the redox potential prediction task, reducing the root mean square error (RMSE) to a level below 0.1 V. Compared with using only the feature descriptors of oxidized or reduced molecules, this method has higher prediction accuracy using molecular feature descriptors with combined modes.

[0034] First, a large amount of structural information of carbonyl compounds is collected and sorted, and some molecules are selected to obtain their redox potential data through DFT calculation. Secondly, based on these molecular structure data, the molecular feature descriptors proposed in the present invention are constructed. Then, the extracted descriptors are trained by machine learning algorithms to establish an efficient and accurate prediction model. Finally, this model is used to predict additional carbonyl compounds, proving that it can quickly and accurately predict their redox potentials, providing an important reference for the design and optimization of liquid flow batteries. The descriptors proposed in the present invention have the advantages of large amount of information, high prediction accuracy, and high calculation efficiency, and can provide strong support for the research and development of carbonyl compounds for liquid flow batteries.

[0035] The present invention proposes a descriptor for predicting the redox potential of carbonyl compounds for flow batteries, which specifically includes the following steps:

[0036] Step 1: Screen a series of carbonyl compound molecules from the PubChem database based on element type, number of rings, and number of heavy atoms. The specific screening criteria are: (1) Contain only H, C, O, N, P, S, F, and Cl elements; (2) The number of molecular rings (a closed system formed by connecting the ends) is less than 3, and for molecules with 0 rings, the number of heavy atoms (non-hydrogen atoms) is less than or equal to 6, for molecules with 1 ring, the number of heavy atoms is less than or equal to 6, for molecules with 2 rings, the number of heavy atoms is less than or equal to 16, and for molecules with 3 rings, the number of heavy atoms is less than or equal to 20; (3) The molecule contains carbonyl groups, and not all of them come from ester groups (for example, the ethyl acetate molecule contains only one carbonyl group from the ester group, which does not meet this condition); (4) It does not contain carboxyl groups or phosphonic acid groups.

[0037] Step 2: Randomly select some of the molecules to calculate the redox potential. The redox potential is calculated based on the difference in Gibbs free energy between the oxidized and reduced molecules. 75% of them are divided as training sets and 25% are divided as test sets.

[0038] Step 3: Based on Mordred descriptor, RDKit descriptor, and Morgan molecular fingerprint and Figure 1 The process in generates the "OH" + "O" molecular descriptor, which is obtained by subtracting the features of the oxidized state molecule from the features of the reduced state molecule in the same dimension. If the result is not 0, it is retained. If the result is 0, the features of the oxidized state or the reduced state molecule are used to replace it. This method takes into account both the differences and similarities between the oxidized and reduced state molecules. For the prediction task of redox potential, it is a more informative molecular feature descriptor.

[0039] Step 4: Use the above descriptors and the oxidized molecular feature descriptors ("O") and reduced molecular feature descriptors ("H") in the machine learning model, and use the Gaussian Process (GP) regression algorithm to train the data in the training set. Use the trained machine learning model to predict the test set, and the predicted values ​​are similar to the calculated values. Figure 2 As shown, the mean square error of the training set is 0.0511V, and the mean square error of the test set is 0.0866V.

[0040] Step 5: Additional evaluation of the accuracy of the model and the effectiveness of the descriptor was performed. First, 978 additional carbonyl molecules that were not in the test set were selected for DFT calculation to obtain redox potentials and the model was used for prediction. The trained machine learning model was then used to predict the redox potentials of 24 carbonyl molecules in the literature and compared with their experimentally measured redox potentials. The results are shown in Figure 2. Figure 3As shown, it can be seen that the “OH” + “O” molecular descriptor achieves the highest accuracy.

[0041] Step 6: Save this machine learning model with the highest accuracy and use it to predict the redox potential of carbonyl molecules later.

Claims

1. A method for predicting redox potential based on carbonyl compound descriptors, characterized in that: The following steps are involved: Step 1: Screen carbonyl compound molecules based on element type, number of rings, and number of heavy atoms; Step 2: Randomly select some of the molecules to calculate the redox potential, and construct a training data set based on the carbonyl compound molecules and their corresponding redox potentials; Step 3: Generate an oxidized state molecular feature descriptor and a reduced state molecular feature descriptor based on the molecular descriptor, and then construct a descriptor for the carbonyl compound according to the difference between the oxidized state molecular feature descriptor and the reduced state molecular feature descriptor; Step 4: Use the newly generated descriptors in the machine learning model and train the data in the training set using the GP regression algorithm; Step 5: The carbonyl compound to be tested is subjected to step 3 to obtain the descriptor, and then the predicted redox potential is obtained through the trained machine learning model.

2. The method for predicting redox potential based on carbonyl compound descriptors according to claim 1, characterized in that: The screening includes: (1) Contains only H, C, O, N, P, S, F, and Cl elements; (2) The number of molecular rings is less than 3. For molecules with 0 rings, the number of heavy atoms other than hydrogen atoms is less than or equal to 6. For molecules with 1 ring, the number of heavy atoms is less than or equal to 6. For molecules with 2 rings, the number of heavy atoms is less than or equal to 16. For molecules with 3 rings, the number of heavy atoms is less than or equal to 20. (3) The molecule contains carbonyl groups, which cannot all come from ester groups; (4) It does not contain carboxyl or phosphonic acid groups.

3. The method for predicting redox potential based on carbonyl compound descriptors according to claim 1, characterized in that: The step 3: generating an oxidized state molecular feature descriptor and a reduced state molecular feature descriptor based on the molecular descriptor, and then constructing a descriptor for carbonyl compounds according to the difference between the oxidized state molecular feature descriptor and the reduced state molecular feature descriptor, as follows: Based on Mordred descriptor, RDKit descriptor, and Morgan molecular fingerprint, the characteristic descriptor of the oxidized state molecule is subtracted from the characteristic descriptor of the reduced state molecule in the same dimension; If the result is not 0, the subtracted feature is used as the descriptor for the carbonyl compound; If the result is 0, the identity of the oxidized or reduced molecule is retained as the descriptor for the carbonyl compound.

4. A redox potential prediction system based on carbonyl compound descriptors, characterized in that: include: The data set construction module is used to screen carbonyl compound molecules according to element type, number of rings, and number of heavy atoms; randomly select some of the molecules to calculate the redox potential, and construct a training data set based on the carbonyl compound molecules and their corresponding redox potentials; A descriptor generation module, used for generating an oxidation state molecular feature descriptor and a reduction state molecular feature descriptor based on the molecular descriptor, and then constructing a descriptor for the carbonyl compound according to the difference between the oxidation state molecular feature descriptor and the reduction state molecular feature descriptor; The model training module is used to use the newly generated descriptors in the machine learning model and train the data in the training set using the GP regression algorithm; The redox potential prediction module is used to obtain the descriptor of the carbonyl compound to be tested through step 3, and then obtain the predicted redox potential through the trained machine learning model.

5. A redox potential prediction device based on carbonyl compound descriptors, characterized in that: It comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement a redox potential prediction method based on a carbonyl compound descriptor as described in any one of claims 1 to 3 when executing the computer program.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting redox potential based on carbonyl compound descriptors as described in any one of claims 1 to 3 is implemented.

Images