Method and device for screening catalysts for carbon dioxide reduction reaction, electronic device and medium

Abstract

The present disclosure relates to a method and device for screening a catalyst for a carbon dioxide reduction reaction, an electronic device and a medium, the method comprising obtaining a screening task, determining reaction process information of each carbon dioxide reduction reaction and geometric structure information of each candidate catalyst; inputting the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst into a prediction model for calculation to obtain adsorption energy of each candidate catalyst-target intermediate combination; determining activity and selectivity of the corresponding candidate catalyst, and screening a target catalyst from a plurality of candidate catalysts according to the activity and selectivity of each candidate catalyst, taking the target catalyst as a screening result of the screening task. The present disclosure can significantly accelerate the screening speed of a new high-performance catalyst suitable for the carbon dioxide reduction reaction, and provide a high-selectivity and high-efficiency solution for industrial applications.

Description

Technical Field

[0001] This disclosure relates to the field of electrochemical reduction catalyst screening technology, and in particular to a method and apparatus, electronic equipment and medium for screening catalysts for carbon dioxide reduction reactions. Background Technology

[0002] The electrochemical reduction of carbon dioxide (CO2) (CO2RR) is considered one of the effective ways to address climate change. Currently, the industrial application of CO2RR is limited by several factors, among which the development of catalysts is crucial. Catalysts directly affect current density, Faraday efficiency, energy efficiency, and stability, thus significantly impacting operating costs.

[0003] However, developing single-atom catalysts with excellent selectivity and activity still faces significant challenges in traditional computational and experimental methods. For example, they are inefficient when screening a large number of candidate catalysts and are computationally expensive. Summary of the Invention

[0004] In view of this, this disclosure proposes a method, apparatus, electronic equipment, and medium for screening catalysts for carbon dioxide reduction reactions, which can significantly accelerate the screening speed of novel high-performance catalysts suitable for carbon dioxide reduction reactions, providing a highly selective and efficient solution for industrial applications.

[0005] According to one aspect of this disclosure, a method for screening catalysts for carbon dioxide reduction reactions is provided, comprising: obtaining a screening task, the screening task indicating a plurality of candidate catalysts and a target reduction reaction, the target reduction reaction including at least one carbon dioxide reduction reaction, and determining reaction process information of each of the carbon dioxide reduction reactions and geometric structure information of each of the candidate catalysts; inputting the reaction process information of each of the carbon dioxide reduction reactions and the geometric structure information of each of the candidate catalysts into a prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination, each candidate catalyst-target intermediate combination being formed based on one candidate catalyst and one target intermediate, all target intermediates being determined based on the target reduction reaction, wherein the prediction model is determined based on a graph neural network; determining the activity and selectivity of the corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and screening a target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, using the target catalyst as the screening result of the screening task.

[0006] Thus, the screening method provided in this disclosure, based on a graph neural network model and an active learning strategy, enables intelligent high-throughput screening, specifically for the efficient screening and design of catalysts used in CO2 reduction reactions. This screening method not only overcomes the shortcomings of traditional screening methods, such as high computational cost and low efficiency, but also achieves automated iterative screening of high-potential catalysts by introducing a predictive model. This method can screen out several highly selective and highly active catalysts, providing the possibility of industrial application for the efficient synthesis of high-value-added products such as methanol. Furthermore, this screening method is versatile, automated, and efficient, applicable not only to the CO2 reduction field but also applicable to the screening and optimization of other electrochemical catalysts, possessing broad industrial prospects and technological value.

[0007] In one possible implementation, the method further includes: acquiring a training sample set; and using the training sample set to iteratively train an initial prediction model to obtain a trained prediction model.

[0008] In one possible implementation, obtaining a training sample set includes: acquiring geometric structure information of multiple sample catalysts and reaction process information of sample reduction reactions; determining each sample catalyst-sample intermediate combination and its corresponding geometric structure information, wherein each sample catalyst-sample intermediate combination is formed based on one sample catalyst and one sample intermediate, and all sample intermediates are determined based on the sample reduction reactions, wherein the sample reduction reactions include at least one carbon dioxide reduction reaction; calculating the adsorption energy of each sample catalyst-sample intermediate combination based on density functional theory; determining multiple training samples, each training sample corresponding to a different sample catalyst-sample intermediate combination, each training sample including the geometric structure information and adsorption energy of the corresponding sample catalyst-sample intermediate combination; and forming the training sample set based on all training samples.

[0009] In this way, for the sample reduction reaction, all possible sample catalyst-sample intermediate combinations, their geometric structure information, and adsorption energies are identified. The model trained based on this information can more accurately predict the interaction between the catalyst and the intermediate, thereby improving the accuracy of the prediction. Furthermore, since the model training covers a variety of different catalyst and intermediate combinations, this helps to improve the generalization ability of the prediction model, enabling it to adapt to carbon dioxide reduction reactions under different conditions.

[0010] In one possible implementation, the initial prediction model is iteratively trained using the training sample set to obtain the trained prediction model, including: in each iteration of training, selecting a subset from the training sample set for this iteration of training; and using the subset to train the current prediction model to obtain the trained prediction model.

[0011] In this way, through iterative training, the predictive model can gradually learn a wider range of sample features, thereby improving its generalization ability on unseen data. Training only with a subset of training samples in each iteration helps optimize the use of computational resources, especially when dealing with large datasets. Compared to processing the entire dataset at once, it reduces the computational burden and accelerates the training process of the predictive model. Furthermore, selecting a new subset for training in each iteration makes the predictive model more flexible and adaptable to changes in data distribution. By continuously learning from different subsets, the predictive model can resist the effects of outliers and noise, enhancing its robustness.

[0012] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: when it is determined that the current iteration of training is in progress, randomly selecting a portion of the training samples from the training sample set to obtain a subset of the training sample set.

[0013] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: if it is determined that the current iteration is not the first iteration of training, using the prediction model obtained from the previous iteration of training to predict the adsorption energy of the sample catalyst-sample intermediate combination corresponding to all training samples in the training sample set, and determining the uncertainty corresponding to each training sample based on the prediction results, and forming the subset based on all training samples whose uncertainty meets the preset conditions.

[0014] In this way, training the model on training samples with high uncertainty in prediction results allows the predictive model to focus more on learning those difficult-to-predict data points, which may improve the overall prediction accuracy of the predictive model. Each iteration selects a new subset based on the current state of the model, dynamically adapting to the learning progress of the predictive model, which helps the predictive model better capture the complexity of the data and reduce the risk of overfitting. With limited computing resources, this training method allows computing resources to be concentrated on the training samples that need to be learned the most, thereby achieving optimal resource allocation. Furthermore, by continuously adjusting the subset used for training, the predictive model can access more diverse data, which helps to enhance the robustness of the predictive model when facing new data.

[0015] In one possible implementation, the activity and selectivity of a corresponding candidate catalyst are determined based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and a target catalyst is screened from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, including: determining the restraint potential of a corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst; determining the activity and selectivity of a corresponding candidate catalyst based on the restraint potential of each candidate catalyst; and screening a target catalyst from the plurality of candidate catalysts according to preset activity screening conditions and selectivity screening conditions, wherein the target catalyst includes at least one candidate catalyst.

[0016] In this way, catalysts with excellent selectivity and activity can be screened out.

[0017] According to another aspect of this disclosure, a screening device for catalysts for carbon dioxide reduction reactions is provided, comprising: an acquisition module for acquiring a screening task, the screening task indicating a plurality of candidate catalysts and a target reduction reaction, the target reduction reaction including at least one carbon dioxide reduction reaction, and determining reaction process information of each of the carbon dioxide reduction reactions and geometric structure information of each of the candidate catalysts; a calculation module for inputting the reaction process information of each of the carbon dioxide reduction reactions and the geometric structure information of each of the candidate catalysts into a prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination, each candidate catalyst-target intermediate combination being formed based on one candidate catalyst and one target intermediate, all target intermediates being determined based on the target reduction reaction, wherein the prediction model is determined based on a graph neural network; and a screening module for determining the activity and selectivity of the corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and screening a target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, using the target catalyst as the screening result of the screening task.

[0018] Thus, the screening device provided in this disclosure, based on a graph neural network model and an active learning strategy, can perform intelligent high-throughput screening, specifically for the efficient screening and design of catalysts used in CO2 reduction reactions. This screening device not only overcomes the shortcomings of traditional screening devices, such as high computational cost and low efficiency, but also achieves automatic iterative screening of high-potential catalysts by introducing a predictive model. This device can screen out several highly selective and highly active catalysts, providing the possibility of industrial application for the efficient synthesis of high-value-added products such as methanol. Furthermore, this screening device possesses versatility, automation, and high efficiency, and is not only applicable to the CO2 reduction field but can also be extended to the screening and optimization of other electrochemical catalysts, possessing broad industrial prospects and technological value.

[0019] In one possible implementation, the apparatus further includes a training module for: acquiring a training sample set; and iteratively training an initial prediction model using the training sample set to obtain a trained prediction model.

[0020] In one possible implementation, obtaining a training sample set includes: acquiring geometric structure information of multiple sample catalysts and reaction process information of sample reduction reactions; determining each sample catalyst-sample intermediate combination and its corresponding geometric structure information, wherein each sample catalyst-sample intermediate combination is formed based on one sample catalyst and one sample intermediate, and all sample intermediates are determined based on the sample reduction reactions, wherein the sample reduction reactions include at least one carbon dioxide reduction reaction; calculating the adsorption energy of each sample catalyst-sample intermediate combination based on density functional theory; determining multiple training samples, each training sample corresponding to a different sample catalyst-sample intermediate combination, each training sample including the geometric structure information and adsorption energy of the corresponding sample catalyst-sample intermediate combination; and forming the training sample set based on all training samples.

[0021] In this way, for the sample reduction reaction, all possible sample catalyst-sample intermediate combinations, their geometric structure information, and adsorption energies are identified. The model trained based on this information can more accurately predict the interaction between the catalyst and the intermediate, thereby improving the accuracy of the prediction. Furthermore, since the model training covers a variety of different catalyst and intermediate combinations, this helps to improve the generalization ability of the prediction model, enabling it to adapt to carbon dioxide reduction reactions under different conditions.

[0022] In one possible implementation, the initial prediction model is iteratively trained using the training sample set to obtain the trained prediction model, including: in each iteration of training, selecting a subset from the training sample set for this iteration of training; and using the subset to train the current prediction model to obtain the trained prediction model.

[0023] In this way, through iterative training, the predictive model can gradually learn a wider range of sample features, thereby improving its generalization ability on unseen data. Training only with a subset of training samples in each iteration helps optimize the use of computational resources, especially when dealing with large datasets. Compared to processing the entire dataset at once, it reduces the computational burden and accelerates the training process of the predictive model. Furthermore, selecting a new subset for training in each iteration makes the predictive model more flexible and adaptable to changes in data distribution. By continuously learning from different subsets, the predictive model can resist the effects of outliers and noise, enhancing its robustness.

[0024] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: when it is determined that the current iteration of training is in progress, randomly selecting a portion of the training samples from the training sample set to obtain a subset of the training sample set.

[0025] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: if it is determined that the current iteration is not the first iteration of training, using the prediction model obtained from the previous iteration of training to predict the adsorption energy of the sample catalyst-sample intermediate combination corresponding to all training samples in the training sample set, and determining the uncertainty corresponding to each training sample based on the prediction results, and forming the subset based on all training samples whose uncertainty meets the preset conditions.

[0026] In this way, training the model on training samples with high uncertainty in prediction results allows the predictive model to focus more on learning those difficult-to-predict data points, which may improve the overall prediction accuracy of the predictive model. Each iteration selects a new subset based on the current state of the model, dynamically adapting to the learning progress of the predictive model, which helps the predictive model better capture the complexity of the data and reduce the risk of overfitting. With limited computing resources, this training method allows computing resources to be concentrated on the training samples that need to be learned the most, thereby achieving optimal resource allocation. Furthermore, by continuously adjusting the subset used for training, the predictive model can access more diverse data, which helps to enhance the robustness of the predictive model when facing new data.

[0027] In one possible implementation, the activity and selectivity of a corresponding candidate catalyst are determined based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and a target catalyst is screened from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, including: determining the restraint potential of a corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst; determining the activity and selectivity of a corresponding candidate catalyst based on the restraint potential of each candidate catalyst; and screening a target catalyst from the plurality of candidate catalysts according to preset activity screening conditions and selectivity screening conditions, wherein the target catalyst includes at least one candidate catalyst.

[0028] In this way, catalysts with excellent selectivity and activity can be screened out.

[0029] According to another aspect of this disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method when executing instructions stored in the memory.

[0030] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided that stores computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.

[0031] According to another aspect of this disclosure, a computer program product is provided, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0032] This disclosure provides a method, apparatus, electronic device, and medium for screening catalysts for carbon dioxide reduction reactions. The method involves acquiring a screening task, which indicates multiple candidate catalysts and target reduction reactions, including at least one carbon dioxide reduction reaction. The method determines the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst. The reaction process information and the geometric structure information of each candidate catalyst are input into a prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination. Each candidate catalyst-target intermediate combination is formed based on one candidate catalyst and one target intermediate. All target intermediates are based on... The target reduction reaction is determined by a prediction model based on a graph neural network. The activity and selectivity of each candidate catalyst are determined by the adsorption energy of all candidate catalyst-target intermediate combinations. Based on the activity and selectivity of each candidate catalyst, the target catalyst is screened from multiple candidate catalysts, and the target catalyst is used as the screening result. This overcomes the shortcomings of high computational cost and low efficiency in traditional screening methods. By introducing a prediction model, automatic iterative screening of high-potential catalysts is achieved, resulting in several highly selective and highly active catalysts. This screening method is universal, automated, and efficient, and has broad industrial prospects and technological value.

[0033] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0034] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0035] Figure 1 A flowchart illustrating a method for screening catalysts for carbon dioxide reduction reactions provided in embodiments of this disclosure is shown.

[0036] Figure 2 This diagram illustrates the EquiformerV2 model and its prediction performance test results provided in an embodiment of this disclosure.

[0037] Figure 3 This diagram illustrates the relationship between vacancy types and inhibition charges in metal and catalyst materials provided in embodiments of this disclosure.

[0038] Figure 4 This diagram illustrates the model iteration results obtained using a screening method according to an embodiment of the present disclosure.

[0039] Figure 5This diagram illustrates a method for screening catalysts for carbon dioxide reduction reactions provided in embodiments of this disclosure.

[0040] Figure 6 A schematic diagram of the simulated catalytic space provided in an embodiment of this disclosure is shown.

[0041] Figure 7 A block diagram of a catalyst screening device for carbon dioxide reduction reaction provided in an embodiment of this disclosure is shown. Detailed Implementation

[0042] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0043] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0044] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0045] To facilitate understanding of the technical solutions provided by the embodiments of this disclosure by those skilled in the art, the technical environment for implementing the technical solutions will be described below.

[0046] With the continued intensification of global warming and the increasing frequency of extreme weather events, the urgency of reducing greenhouse gas emissions has become increasingly apparent. Current global temperatures have risen by 1.2°C above pre-industrial levels, and countries have reached a consensus in the Paris Agreement to immediately implement emission reduction measures to prevent irreversible damage to ecological and socio-economic systems. Against this backdrop, the electrochemical reduction of carbon dioxide (CO2) (CO2RR) is considered one of the effective ways to address climate change. CO2RR can convert CO2 into high-value-added chemicals and renewable fuels, providing a feasible path for industrial decarbonization and future carbon neutrality. Although CO2 reduction to CO has made some progress in industry, there is still great potential for further conversion into other high-value-added products.

[0047] Currently, the industrial application of CO2RR is limited by techno-economic factors, with multiple aspects such as catalyst design, system design, and electrolyzer design affecting its economic viability. Catalyst development, in particular, is a key factor, directly influencing current density, Faradaic efficiency, energy efficiency, and stability, thus significantly impacting operating costs. However, highly selective and highly active CO2RR catalysts are currently scarce, especially for the reduction of CO2 to high-value-added chemicals such as methanol.

[0048] However, developing catalysts with excellent selectivity and activity remains a significant challenge using traditional computational and experimental methods. First, traditional methods are inefficient and computationally expensive when screening large numbers of potential catalysts. Second, many studies employ feature-based designs based on specific structures, such as alloy nanoparticles, metal oxides, and metal two-site catalysts. While this feature-based design, relying on human expertise, enhances physical understanding, it limits the transferability of models across different systems and necessitates expert redesign, thus limiting its applicability. Therefore, developing an efficient method for exploring a broad catalyst design space and designing single-atom catalysts has become an urgent need in this field.

[0049] To address the aforementioned technical problems, this disclosure provides a method for screening catalysts for carbon dioxide reduction reactions. Now, in conjunction with... Figures 1 to 6 The method for screening catalysts for carbon dioxide reduction reactions provided in the embodiments of this disclosure is illustrated.

[0050] like Figure 1 As shown, the screening method may include the following steps S101 to S103.

[0051] Step S101: Obtain the screening task. The screening task is used to indicate multiple candidate catalysts and target reduction reactions. The target reduction reaction includes at least one carbon dioxide reduction reaction. The reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst are determined.

[0052] The target reduction reaction is a type of reduction reaction of interest, and the number of specific carbon dioxide reduction reactions included in it is set according to actual needs.

[0053] In some embodiments, the candidate catalyst can be a single-atom catalyst (SAC). Correspondingly, the screened target catalyst is also a single-atom catalyst. Among them, single-atom catalysts have shown great potential in CO2RR catalysis due to their high atomic utilization, low number of coordinated metal atoms, unique electronic structure, and strong metal-support interaction. Because single-atom catalysts possess advantages such as low cost, tunable structure, high energy efficiency, and low overpotential, they are considered ideal candidates for next-generation CO2RR catalysts. Both the candidate catalyst and the target catalyst described below are essentially catalysts.

[0054] Step S102: Input the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst into the prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination. Each candidate catalyst-target intermediate combination is formed based on one candidate catalyst and one target intermediate. All target intermediates are determined based on the target reduction reaction. The prediction model is determined based on a graph neural network.

[0055] The prediction model is a machine learning model based on a graph neural network (GNN) used to predict adsorption energies. Unlike traditional machine learning models that rely on manually designed features, the prediction model provided in this disclosure transforms the catalyst's geometry into a graph structure, enabling direct modeling of interatomic interactions and improving adaptability to different catalyst types. Furthermore, the prediction model provided in this disclosure is obtained through pre-training on a large-scale adsorption energy dataset and fine-tuning on a specific catalyst dataset. This allows the prediction model to accurately predict the catalyst's adsorption energy, significantly improving screening efficiency. Details of the prediction model's training method are provided below.

[0056] Step S103: Determine the activity and selectivity of the corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and screen the target catalyst from multiple candidate catalysts based on the activity and selectivity of each candidate catalyst, and use the target catalyst as the screening result of the screening task.

[0057] To better facilitate the target reduction reaction, this screening method focuses on two key indicators: catalyst activity and selectivity. Based on these two indicators, ideal target catalysts are selected. Catalyst selectivity refers to the degree to which a catalyst promotes different reactions in a reaction system capable of multiple reactions. Catalyst activity, or catalytic activity, refers to the catalyst's ability to catalyze.

[0058] Thus, the screening method provided in this disclosure, based on a graph neural network model and an active learning strategy, enables intelligent high-throughput screening, specifically for the efficient screening and design of catalysts used in CO2 reduction reactions. This screening method not only overcomes the shortcomings of traditional screening methods, such as high computational cost and low efficiency, but also achieves automated iterative screening of high-potential catalysts by introducing a predictive model. This method can screen out several highly selective and highly active catalysts, providing the possibility of industrial application for the efficient synthesis of high-value-added products such as methanol. Furthermore, this screening method is versatile, automated, and efficient, applicable not only to the CO2 reduction field but also applicable to the screening and optimization of other electrochemical catalysts, possessing broad industrial prospects and technological value.

[0059] The screening method provided in this disclosure employs an automatically iterative active learning framework to achieve efficient exploration and data optimization during catalyst screening. The active learning framework selects the candidate catalysts with the highest uncertainty and greatest screening potential for a new round of model training through iterative data sampling and model updates. Unlike traditional screening methods, this active learning framework can gradually improve model accuracy during the learning cycle while avoiding computation on a large number of invalid samples, thus achieving optimal resource utilization.

[0060] This screening method may also include a training process for the prediction model, which may include: obtaining a training sample set; iteratively training the initial prediction model using the training sample set to obtain the trained prediction model.

[0061] During the training of the prediction model, obtaining the training sample set may include: acquiring the geometric structure information of multiple sample catalysts and the reaction process information of sample reduction reactions; determining the catalyst-intermediate combination and corresponding geometric structure information of each sample, where each catalyst-intermediate combination is formed based on one sample catalyst and one sample intermediate, and all sample intermediates are determined based on the sample reduction reactions, wherein the sample reduction reactions include at least one carbon dioxide reduction reaction; calculating the adsorption energy of each catalyst-intermediate combination based on density functional theory (DFT); determining multiple training samples, each training sample corresponding to a different catalyst-intermediate combination, and each training sample including the geometric structure information and adsorption energy of the corresponding catalyst-intermediate combination; and forming a training sample set based on all training samples. In this way, for the sample reduction reaction, all possible sample catalyst-sample intermediate combinations, their geometric structure information, and adsorption energies are identified. The model trained based on this information can more accurately predict the interaction between the catalyst and the intermediate, thereby improving the accuracy of the prediction. Furthermore, since the model training covers a variety of different catalyst and intermediate combinations, this helps to improve the generalization ability of the prediction model, enabling it to adapt to carbon dioxide reduction reactions under different conditions.

[0062] In some embodiments, a large number of candidate catalysts can be screened using a DFT computational platform. This platform can estimate the activity and selectivity of candidate catalysts based on simulation software such as VASP and employing a computational hydrogen electrode (CHE) model. This allows for rapid determination of catalyst activity and selectivity for the CO2 reduction reaction (CO2RR), ensuring the accuracy of the initial screening and selecting candidate catalysts with superior activity and selectivity as the data basis for subsequently building the training sample set. This provides high-quality data for the subsequent training of the prediction model. In the DFT calculation, the Projected Augmented Wave (PAW) method and commonly used functionals such as the PBE (Perdew-Burke-Ernzerhof) functional are used to calculate the electron exchange energy, and DFT-D3 correction is incorporated to describe mid- and long-range interactions. All calculations consider spin polarization to ensure the accuracy of the electronic structure. Furthermore, to avoid the influence of periodic boundary conditions, a 15 Å boundary value can be used. The vacuum layer isolates the adsorption surface, making the simulated environment of each candidate catalyst closer to the real experimental conditions, so as to obtain a more realistic adsorption energy.

[0063] Traditional catalyst screening methods often suffer from low efficiency and high computational resource consumption. This screening method, however, significantly improves efficiency through a high-throughput DFT computing platform. The DFT platform automates the processing, analysis, and recording of the adsorption energy of each candidate catalyst-sample intermediate combination, ensuring the accuracy and consistency of the screening and providing reliable basic data for further analysis.

[0064] During the training process of the prediction model, the initial prediction model is iteratively trained using a training sample set to obtain the trained prediction model. This may include: in each iteration, selecting a subset from the training sample set for that iteration; and using this subset to train the current prediction model, resulting in the trained prediction model. Through iterative training, the prediction model can gradually learn a wider range of sample features, thereby improving its generalization ability on unseen data. Training with only a subset of training samples in each iteration helps optimize the use of computational resources, especially when dealing with large-scale datasets. Compared to processing the entire dataset at once, it reduces the computational burden and accelerates the training process of the prediction model. Furthermore, selecting a new subset for training in each iteration makes the prediction model more flexible and adaptable to changes in data distribution. By continuously learning from different subsets, the prediction model can resist the influence of outliers and noise, enhancing its robustness.

[0065] Selecting a subset from the training sample set for this iteration of training may include: during the first iteration of training, randomly selecting a portion of the training samples from the training sample set, using this portion as a subset. Selecting a subset from the training sample set for this iteration of training may also include: during subsequent iterations of training, using the prediction model obtained from the previous iteration to predict the adsorption energy of the sample catalyst-sample intermediate combination corresponding to all training samples in the training sample set, and determining the uncertainty corresponding to each training sample based on the prediction results; then forming a subset based on all training samples whose uncertainties satisfy preset conditions. In this way, training the model on training samples with high uncertainty in prediction results allows the predictive model to focus more on learning those difficult-to-predict data points, which may improve the overall prediction accuracy of the predictive model. Each iteration selects a new subset based on the current state of the model, dynamically adapting to the learning progress of the predictive model, which helps the predictive model better capture the complexity of the data and reduce the risk of overfitting. With limited computing resources, this training method allows computing resources to be concentrated on the training samples that need to be learned the most, thereby achieving optimal resource allocation. Furthermore, by continuously adjusting the subset used for training, the predictive model can access more diverse data, which helps to enhance the robustness of the predictive model when facing new data.

[0066] The screening method provided in this disclosure employs an active learning strategy, achieving efficient screening and data optimization through automatic iterative sampling and model updates. For example, the active learning framework starts from a complete training sample set containing 100 training samples. First, it randomly selects several training samples, such as eight, and forms a subset based on these eight samples for the prediction model's first iteration of training. Then, using five-fold cross-validation, it trains and evaluates the initial prediction model using this subset, obtaining the prediction model after the first iteration. Next, it uses this prediction model to predict the adsorption energy of the catalyst-intermediate combination corresponding to the 100 training samples in the complete training sample set, and determines the adsorption energy based on the prediction results (i.e., adsorption energy) of these 100 training samples. The adsorption energies of the 100 training samples each correspond to uncertainties. In this example, these uncertainties can be estimated through variance. If the uncertainties of 10 training samples meet preset conditions (e.g., exceeding a preset threshold), then these 10 training samples form a subset for the second iteration of training. In each subsequent iteration (second, third, etc.), the active learning framework automatically selects several training samples with the highest uncertainties to train the prediction model, thereby improving the prediction accuracy. Compared to processing the entire dataset at once, this reduces the computational burden of each iteration, thus accelerating the training process of the prediction model. The iterative process of the active learning framework provided in this embodiment significantly improves screening efficiency. Simultaneously, the active learning framework saves the calculated and screened data in each iteration for cross-validation of future experimental and simulation results. The active learning process of this embodiment can adapt to different catalyst screening needs through continuous optimization, providing new ideas for high-throughput screening tasks in other fields.

[0067] During the data sampling process, the screening method provided in this embodiment employs a multi-dimensional hybrid sampling strategy. Each iteration not only selects high-performing sample catalyst-sample intermediate combinations for verification but also randomly extracts samples from globally unexplored areas to ensure the diversity of the dataset and the comprehensiveness of the model. After each round of sampling, new data points are automatically added to the training set, updating the prediction model and gradually improving the accuracy of the prediction model in adsorption energy prediction.

[0068] During the training of the predictive model, the embodiments of this disclosure can achieve automated high-throughput DFT calculation and data processing, greatly improving the speed and efficiency of catalyst screening. By combining a self-developed code package, the calculation data of each round can be automatically collected, cleaned, and stored, reducing errors caused by human intervention. In addition, this screening method performs multiple rounds of screening for the candidate catalyst materials, providing a solid data foundation for machine learning prediction.

[0069] In the specific implementation, the EquiformerV2 model (such as...) is adopted.Figure 2 As shown in Figure a, this model utilizes an embedding layer to perform node and edge embedding on the input geometric structure information to obtain corresponding node and edge features. These node and edge features are then input into an attention module containing attention units and feedforward units for computation, yielding the adsorption energy corresponding to the geometric structure information. This model contains 31 million parameters and is pre-trained on the OC20+22 dataset. The OC20+22 dataset includes adsorption energy prediction tasks for various material systems, demonstrating strong adaptability and good transferability. During fine-tuning, the learning rate is reduced to adjust all parameters, allowing the prediction model to focus on specific CO2RR reaction systems. This fine-tuning method retains the original model's strong adaptability in generalization tasks while precisely optimizing for the target reduction reaction of interest, thereby enhancing the accuracy of adsorption energy prediction. The performance test results of the prediction model can be obtained through... Figure 2 Figure b shows the changes in the prediction error distribution and mean absolute error (MAE) with the number of iterations. Figure 2 Figure c shows the statistic Ri, calculated based on the predicted adsorption energy and the DFT-calculated adsorption energy, which measures the goodness of fit of the prediction model. 2 The Root Mean Square Error (RMSE) and MAE indicate that the prediction model has a good fit and low error.

[0070] This screening method employs five-fold cross-validation to train and validate the prediction model, ensuring the robustness and accuracy of the prediction results. In each iteration, the average of the predictions is used to predict the adsorption energy, while the variance is used to estimate the prediction uncertainty. This prediction model not only provides rapid and accurate activity assessment of the candidate catalyst but also exhibits strong scalability, making it applicable to other similar electrochemical catalytic reaction systems. Ultimately, the optimal training samples are obtained through the prediction model, providing a foundation for sampling in the subsequent active learning process.

[0071] Thus, this screening method, by combining high-throughput DFT calculations with a graph neural network model, achieves rapid screening of a large number of single-atom catalysts, overcoming the problems of high computational resource consumption and low screening efficiency in traditional screening methods. By combining deep learning and active learning, an automatically iterative screening framework is designed to optimize the prediction of catalyst activity and selectivity, and selects the most promising catalyst-intermediate combination for calculation based on uncertainty, thereby significantly improving the speed and accuracy of screening. Furthermore, the graph neural network model used in this screening method does not rely on manual feature design, possessing high adaptability and versatility, allowing the screening process to be easily extended to different material systems. Simultaneously, the active learning framework continuously updates and optimizes the screening strategy through multiple rounds of learning iterations, effectively improving the screening accuracy and resource utilization efficiency of candidate materials.

[0072] like Figure 3 The relationship between different metals, vacancy types in different catalyst materials, and limiting potentials determined based on adsorption energy can be used to determine the activity and selectivity of catalysts through limiting potentials, thereby enabling catalyst screening. In some embodiments, step S103, which involves determining the activity and selectivity of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and then screening a target catalyst from multiple candidate catalysts based on the activity and selectivity of each candidate catalyst, may include: determining the restraining potential of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst. For example, the restraining potential of candidate catalyst A can be obtained based on the adsorption energies of all combinations corresponding to candidate catalyst A (i.e., candidate catalyst A-target intermediate 1 combination, candidate catalyst A-target intermediate 2 combination, ..., candidate catalyst A-target intermediate N combination). The method for determining the restraining potential can be flexibly selected according to actual needs, and this disclosure does not limit it; determining the activity and selectivity of a candidate catalyst based on the restraining potential of each candidate catalyst; and screening a target catalyst from multiple candidate catalysts according to preset activity screening conditions and selectivity screening conditions. The target catalyst includes at least one candidate catalyst. This screening method can yield... Figure 4 The iterative results shown Figure 4 The horizontal axis represents the restraint potential of each catalyst, and the vertical axis represents the selectivity of various catalysts. Through several rounds of iteration, the screening results can yield catalysts with significantly improved performance compared to existing CuN4 catalysts, such as the Pd-C3P catalyst based on palladium (Pd), carbon (C), and phosphorus (P).

[0073] In some embodiments, in addition to adsorption energy, the binding energy of each candidate catalyst can also be used for screening to obtain a target catalyst that has a relatively good balance of stability, activity and selectivity.

[0074] The intelligent screening method based on density functional theory calculations, graph neural networks, and active learning strategies provided in this disclosure enables high-efficiency and high-precision catalyst screening. It overcomes the adaptability issues faced by traditional methods in designing general models. Active learning technology offers a new approach to solving these problems. By employing an automatically iterative machine learning framework, it not only reduces the predictive uncertainty of the prediction model during screening but also significantly accelerates the screening speed of novel high-performance CO2RR catalysts, providing a highly selective and efficient solution for industrial applications. This screening method overcomes the efficiency and applicability limitations of traditional screening methods. By combining advanced deep learning models and DFT calculations, it achieves efficient catalyst screening and evaluation, promoting the generation of high-value-added chemicals and meeting the needs of industrial decarbonization applications in the electrochemical field.

[0075] This screening method innovatively utilizes a graph neural network model, which is pre-trained and then fine-tuned for a specific adsorption energy database. Compared to models that rely on traditional manually designed features, graph neural networks exhibit extremely high adaptability and versatility in handling different types of catalyst materials, eliminating the need for structural feature redesign during the screening process. Thus, the prediction model provided in this disclosure can not only quickly predict adsorption energies to determine the corresponding catalyst activity and selectivity, but also supports the screening of multiple catalyst material types, greatly improving the model's transferability and versatility.

[0076] This screening method combines large-scale DFT computation with deep learning prediction in a high-throughput screening process. It boasts a high degree of automation, simplifies the operation, and makes it easy for non-expert users to learn. This screening method employs a modular design, such as... Figure 5 As shown, the catalytic space can be simulated simply by inputting the catalyst screening space and the catalyst reaction space (e.g., Figure 6This method constructs a training sample set using the DFT computation platform within an active learning framework, followed by graph neural network training and rapid catalyst prediction. Density functional theory verification is then performed, utilizing the adsorption energy calculated by DFT to tune the prediction model. Active learning sampling is then performed even if the prediction model has not converged; once converged, the trained prediction model is obtained, thus predicting the optimal selection result for the target catalyst. Furthermore, through continuous optimization of the prediction model, the screening process can adapt to new catalyst design requirements, exhibiting high scalability and practicality. The organic combination of high-throughput DFT computation, graph neural network models, and active learning strategies not only significantly improves the speed and accuracy of CO2RR catalyst screening but also provides a convenient and practical solution for the screening and development of novel catalysts. This screening method can be widely applied to the theoretical design and screening of catalysts for other reactions, possessing high promotional value and industrialization prospects.

[0077] This disclosure also provides a screening device for catalysts used in carbon dioxide reduction reactions, comprising: an acquisition module for acquiring a screening task, the screening task indicating a plurality of candidate catalysts and a target reduction reaction, the target reduction reaction including at least one carbon dioxide reduction reaction, and determining the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst; a calculation module for inputting the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst into a prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination, each candidate catalyst-target intermediate combination being formed based on one candidate catalyst and one target intermediate, all target intermediates being determined based on the target reduction reaction, wherein the prediction model is determined based on a graph neural network; and a screening module for determining the activity and selectivity of the corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and screening the target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, using the target catalyst as the screening result of the screening task.

[0078] Thus, the screening device provided in this disclosure, based on a graph neural network model and an active learning strategy, can perform intelligent high-throughput screening, specifically for the efficient screening and design of catalysts used in CO2 reduction reactions. This screening device not only overcomes the shortcomings of traditional screening devices, such as high computational cost and low efficiency, but also achieves automatic iterative screening of high-potential catalysts by introducing a predictive model. This device can screen out several highly selective and highly active catalysts, providing the possibility of industrial application for the efficient synthesis of high-value-added products such as methanol. Furthermore, this screening device possesses versatility, automation, and high efficiency, and is not only applicable to the CO2 reduction field but can also be extended to the screening and optimization of other electrochemical catalysts, possessing broad industrial prospects and technological value.

[0079] In one possible implementation, the apparatus further includes a training module for: acquiring a training sample set; and iteratively training an initial prediction model using the training sample set to obtain a trained prediction model.

[0080] In one possible implementation, obtaining a training sample set includes: acquiring geometric structure information of multiple sample catalysts and reaction process information of sample reduction reactions; determining each sample catalyst-sample intermediate combination and its corresponding geometric structure information, wherein each sample catalyst-sample intermediate combination is formed based on one sample catalyst and one sample intermediate, and all sample intermediates are determined based on the sample reduction reaction, wherein the sample reduction reaction includes at least one carbon dioxide reduction reaction; calculating the adsorption energy of each sample catalyst-sample intermediate combination based on density functional theory; determining multiple training samples, each training sample corresponding to a different sample catalyst-sample intermediate combination, and each training sample including the geometric structure information and adsorption energy of the corresponding sample catalyst-sample intermediate combination; and forming the training sample set based on all training samples. In this way, for sample reduction reactions, all possible sample catalyst-sample intermediate combinations and their geometric structure information and adsorption energies are determined. The model trained based on this information can more accurately predict the interaction between the catalyst and the intermediate, thereby improving the accuracy of the prediction. Furthermore, since the model training covers a variety of different catalyst and intermediate combinations, this helps to improve the generalization ability of the prediction model, enabling it to adapt to carbon dioxide reduction reactions under different conditions.

[0081] In one possible implementation, the initial prediction model is iteratively trained using the training sample set to obtain the trained prediction model. This includes: in each iteration, selecting a subset from the training sample set for the current iteration; and training the current prediction model using the subset to obtain the trained prediction model. Through iterative training, the prediction model can gradually learn a wider range of sample features, thereby improving its generalization ability on unseen data. Training with only a subset of training samples in each iteration helps optimize the use of computational resources, especially when dealing with large datasets. Compared to processing the entire dataset at once, this reduces the computational burden and accelerates the training process of the prediction model. Furthermore, selecting a new subset for training in each iteration makes the prediction model more flexible and adaptable to changes in data distribution. By continuously learning from different subsets, the prediction model can resist the influence of outliers and noise, enhancing its robustness.

[0082] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: when it is determined that the current iteration of training is in progress, randomly selecting a portion of the training samples from the training sample set to obtain a subset of the training sample set.

[0083] In one possible implementation, selecting a subset from the training sample set for this iteration of training includes: if it is determined that the current iteration is not the first iteration of training, using the prediction model obtained from the previous iteration of training to predict the adsorption energy of the sample catalyst-sample intermediate combination corresponding to all training samples in the training sample set, and determining the uncertainty corresponding to each training sample based on the prediction results, and forming the subset based on all training samples whose uncertainty meets preset conditions. In this way, training the model based on training samples with higher prediction uncertainty allows the prediction model to focus more on learning those difficult-to-predict data points, potentially improving the overall prediction accuracy of the prediction model; each iteration selects a new subset based on the current state of the model, dynamically adapting to the learning progress of the prediction model, which helps the prediction model better capture the complexity of the data and reduce the risk of overfitting; with limited computing resources, this training method allows computing resources to be concentrated on the training samples that most need learning, thereby achieving optimal resource allocation; and by continuously adjusting the subset used for training, the prediction model can access more diverse data, which helps enhance the robustness of the prediction model when facing new data.

[0084] In one possible implementation, the activity and selectivity of a candidate catalyst are determined based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and a target catalyst is screened from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst. This includes: determining the restraint potential of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst; determining the activity and selectivity of a candidate catalyst based on the restraint potential of each candidate catalyst; and screening a target catalyst from the plurality of candidate catalysts according to preset activity and selectivity screening conditions, wherein the target catalyst includes at least one candidate catalyst. In this way, catalysts with excellent selectivity and activity can be screened.

[0085] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0086] This disclosure also proposes a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method. The computer-readable storage medium can be volatile or non-volatile.

[0087] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0088] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0089] Figure 7 A block diagram of a catalyst screening device for carbon dioxide reduction reactions provided in an embodiment of this disclosure is shown. For example, device 1900 can be provided as a server or terminal device. (Refer to...) Figure 7 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0090] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0091] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0092] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0093] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0094] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0095] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0096] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0097] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0098] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0099] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0100] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for screening catalysts for carbon dioxide reduction reactions, characterized in that, include: A screening task is obtained, which is used to indicate multiple candidate catalysts and target reduction reactions, the target reduction reactions including at least one carbon dioxide reduction reaction, and to determine the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst. The reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst are input into the prediction model for calculation to obtain the adsorption energy of each candidate catalyst-target intermediate combination. Each candidate catalyst-target intermediate combination is formed based on one candidate catalyst and one target intermediate. All target intermediates are determined based on the target reduction reaction. The prediction model is determined based on a graph neural network. The prediction model includes the EquiformerV2 model, which is obtained by pre-training based on the training sample set and fine-tuning the full parameters to focus the model on the carbon dioxide reduction reaction system. The prediction model is trained by an active learning strategy for automatic iterative sampling and model update. The active learning strategy includes randomly selecting a subset from the training sample set as the subset used for the first iteration training, using the multiple training samples with the highest uncertainty obtained from the previous prediction as the subset used for non-first iterations, and combining the adsorption energy calculated by high-throughput density functional theory for model parameter tuning. The activity and selectivity of the corresponding candidate catalyst are determined based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and the target catalyst is screened from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, and the target catalyst is used as the screening result of the screening task. The process involves determining the activity and selectivity of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and then screening a target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst. This includes: determining the restraining potential of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst; determining the activity and selectivity of a candidate catalyst based on the restraining potential of each candidate catalyst; and screening a target catalyst from the plurality of candidate catalysts according to preset activity and selectivity screening conditions. The target catalyst includes at least one candidate catalyst.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the training sample set; The initial prediction model is iteratively trained using the training sample set to obtain the trained prediction model.

3. The method according to claim 2, characterized in that, Obtain the training sample set, including: Geometric structure information of multiple sample catalysts and reaction process information of sample reduction reactions are obtained to determine each sample catalyst-sample intermediate combination and its corresponding geometric structure information. Each sample catalyst-sample intermediate combination is formed based on one sample catalyst and one sample intermediate. All sample intermediates are determined based on the sample reduction reactions, wherein the sample reduction reactions include at least one carbon dioxide reduction reaction. The adsorption energy of each sample catalyst-sample intermediate combination was calculated based on density functional theory. Multiple training samples were identified, each training sample corresponding to a different sample catalyst-sample intermediate combination, and each training sample included the geometric structure information and adsorption energy of the corresponding sample catalyst-sample intermediate combination; The training sample set is formed based on all training samples.

4. The method according to claim 2 or 3, characterized in that, The initial prediction model is iteratively trained using the training sample set to obtain the trained prediction model, including: during each iteration of training, Select a subset from the training sample set for this iteration of training; The current prediction model is trained using the subset to obtain the trained prediction model.

5. The method according to claim 4, characterized in that, A subset for this iteration of training is selected from the training sample set, including: Given that we are currently in the first iteration of training, we randomly select a subset of training samples from the training sample set to obtain a subset of the training sample set.

6. The method according to claim 4, characterized in that, A subset for this iteration of training is selected from the training sample set, including: If it is determined that the current training is not the first iteration, the adsorption energy of the sample catalyst-sample intermediate combination corresponding to all training samples in the training sample set is predicted using the prediction model obtained from the previous iteration training. The uncertainty corresponding to each training sample is determined based on the prediction results. All training samples that meet the preset conditions based on the uncertainty are formed into the subset.

7. A screening device for a catalyst in a carbon dioxide reduction reaction, characterized in that, include: An acquisition module is used to acquire a screening task, which indicates multiple candidate catalysts and target reduction reactions, the target reduction reactions including at least one carbon dioxide reduction reaction, and to determine the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst. The calculation module is used to input the reaction process information of each carbon dioxide reduction reaction and the geometric structure information of each candidate catalyst into the prediction model for calculation, so as to obtain the adsorption energy of each candidate catalyst-target intermediate combination. Each candidate catalyst-target intermediate combination is formed based on one candidate catalyst and one target intermediate. All target intermediates are determined based on the target reduction reaction. The prediction model is determined based on a graph neural network. The prediction model includes the EquiformerV2 model, which is obtained by pre-training based on the training sample set and fine-tuning the full parameters to make the model focus on the carbon dioxide reduction reaction system. The prediction model is trained by an active learning strategy for automatic iterative sampling and model update. The active learning strategy includes randomly selecting a subset from the training sample set as the subset used for the first iteration training, using the multiple training samples with the highest uncertainty obtained from the previous prediction as the subset used for non-first iterations, and combining the adsorption energy calculated by high-throughput density functional theory for model parameter tuning. The screening module is used to determine the activity and selectivity of the corresponding candidate catalyst based on the adsorption energy of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and to screen the target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst, and to use the target catalyst as the screening result of the screening task. The process involves determining the activity and selectivity of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst, and then screening a target catalyst from the plurality of candidate catalysts based on the activity and selectivity of each candidate catalyst. This includes: determining the restraining potential of a candidate catalyst based on the adsorption energies of all candidate catalyst-target intermediate combinations corresponding to each candidate catalyst; determining the activity and selectivity of a candidate catalyst based on the restraining potential of each candidate catalyst; and screening a target catalyst from the plurality of candidate catalysts according to preset activity and selectivity screening conditions. The target catalyst includes at least one candidate catalyst.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the method of any one of claims 1 to 6 when executing instructions stored in the memory.

9. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images