Method and apparatus for constructing catalyst system relaxation energy prediction model

Abstract

A method for constructing a catalyst system relaxation energy prediction model, executed by a computing device, comprising: acquiring a first training data set, the first training data set comprising a plurality of pieces of first training data, and each piece of first training data comprising a first training sample and a corresponding first sample label, wherein the first training sample comprises first catalyst system structure information, and the first sample label comprises system energy information corresponding to the first catalyst system structure information; training a catalyst system energy prediction model by using the first training data set, so as to obtain a pre-trained catalyst system energy prediction model; on the basis of the pre-trained catalyst system energy prediction model, constructing a catalyst system relaxation energy initial prediction model; acquiring a second training data set, the second training data set comprising a plurality of pieces of second training data, and each piece of second training data comprising a second training sample and a corresponding second sample label, wherein the second training sample comprises second catalyst system structure information, and the second sample label comprises relaxation energy information corresponding to the second catalyst system structure information; and training the catalyst system relaxation energy initial prediction model by using the second training data set, so as to obtain a catalyst system relaxation energy prediction model.

Description

Method and device for constructing catalyst system relaxation energy prediction model

[0001] Related applications

[0002] This application claims priority to Chinese patent application number 2023111904865, filed on September 14, 2023, entitled “Method and device for constructing a relaxation energy prediction model for a catalyst system”, and priority to Chinese patent application number 2023111901833, filed on September 14, 2023, entitled “Method and device for constructing a relaxation energy prediction model for a catalyst system”, the entire text of which is hereby incorporated by reference. Technical Field

[0003] The present application relates to the field of computer technology, and in particular to a method and apparatus for constructing a catalyst system relaxation energy prediction model, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0004] Catalysts are important auxiliary means of chemical reactions and are widely used in the fields of chemical industry and manufacturing. Selecting a suitable catalyst for a chemical reaction is an important means to control the speed of the chemical reaction and ensure the safety of the reaction. In traditional methods, a large number of experiments are generally used to select suitable catalysts from a large number of alternative substances. This method is not only time-consuming and labor-intensive, but also causes great waste. To this end, experimentalists hope to predict some key indicators of catalysts in advance to reduce the number of alternatives for catalyst screening and to quickly discover suitable catalysts. Relaxation energy can indicate the energy released by the adsorbate-catalyst system from the initial state to the stable state, and is a key indicator for measuring catalyst performance. However, the traditional relaxation energy prediction method based on quantum mechanics has high computational complexity and huge amount of calculation, which seriously affects the efficiency of catalyst research and development. In recent years, with the development of computer technology, the use of deep learning methods to construct relaxation energy prediction models to predict the relaxation energy of adsorbate-catalyst systems has received increasing attention.

[0005] Summary of the Invention

[0006] The present application provides a method and apparatus for constructing a catalyst system relaxation energy prediction model, a computing device, a computer-readable storage medium, and a computer program product.

[0007] According to one aspect of the present application, a method for constructing a catalyst system relaxation energy prediction model is provided, which is executed by a computing device, and the method includes:

[0008] Obtaining a first training data set, the first training data set comprising a plurality of first training data, each first training data comprising a first training sample and a corresponding first sample label, wherein the first training sample comprises first catalyst system structure information, and the first sample label comprises system energy information corresponding to the first catalyst system structure information;

[0009] Training a catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model;

[0010] Constructing an initial catalyst system relaxation energy prediction model based on the pre-trained catalyst system energy prediction model;

[0011] Obtaining a second training data set, the second training data set comprising a plurality of second training data, each second training data comprising a second training sample and a corresponding second sample label, wherein the second training sample comprises second catalyst system structure information, and the second sample label comprises relaxation energy information corresponding to the second catalyst system structure information; and

[0012] The catalyst system relaxation energy initial prediction model is trained using the second training data set to obtain a catalyst system relaxation energy prediction model.

[0013] According to another aspect of the present application, a catalyst system relaxation energy prediction model construction device is provided, comprising:

[0014] a first acquisition module, configured to acquire a first training data set, the first training data set comprising a plurality of first training data, each first training data comprising a first training sample and a corresponding first sample label, wherein the first training sample comprises first catalyst system structure information, and the first sample label comprises system energy information corresponding to the first catalyst system structure information;

[0015] a first training module, configured to train a catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model;

[0016] A first construction module is used to construct an initial prediction model of catalyst system relaxation energy based on the pre-trained catalyst system energy prediction model;

[0017] a second acquisition module, configured to acquire a second training data set, the second training data set comprising a plurality of second training data, each second training data comprising a second training sample and a corresponding second sample label, wherein the second training sample comprises second catalyst system structure information, and the second sample label comprises relaxation energy information corresponding to the second catalyst system structure information;

[0018] The second training module is used to train the catalyst system relaxation energy initial prediction model using the second training data set to obtain the catalyst system relaxation energy prediction model.

[0019] According to another aspect of the present application, a method for predicting the relaxation energy of a catalyst system is provided, comprising: obtaining the system structure of the catalyst system to be predicted; inputting the system structure of the catalyst system to be predicted into the catalyst system relaxation energy prediction model in any of the aforementioned embodiments to obtain the relaxation energy of the catalyst system to be predicted.

[0020] According to another aspect of the present application, a computing device is provided, comprising: a memory configured to store computer-executable instructions; and a processor configured to perform the steps of the catalyst system relaxation energy prediction model construction method according to some embodiments of the present application when the computer-executable instructions are executed by the processor.

[0021] According to another aspect of the present application, a computer-readable storage medium is provided, which stores computer-executable instructions. When the computer-executable instructions are executed, they implement the steps of the catalyst system relaxation energy prediction model construction method according to some embodiments of the present application.

[0022] According to another aspect of the present application, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the catalyst system relaxation energy prediction model construction method according to some embodiments of the present application.

[0023] These and other advantages of the application will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the following briefly introduces the drawings required for use in the embodiments or the conventional technology descriptions. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the disclosed drawings without any creative work.

[0025] FIG1 is a schematic diagram showing the catalyst development process in the examples of the present application;

[0026] FIG2 shows an exemplary application scenario of the method for constructing a catalyst system relaxation energy prediction model in an embodiment of the present application;

[0027] FIG3 shows a schematic flow chart of a method for constructing a catalyst system relaxation energy prediction model in an embodiment of the present application;

[0028] FIG4 shows a catalyst system energy prediction model trained using a first training data set in an embodiment of the present application to obtain a pre-trained catalyst system energy prediction model;

[0029] FIG5 shows a schematic diagram of training a catalyst system energy prediction model in an embodiment of the present application;

[0030] FIG6 shows a schematic flow chart of training the catalyst system relaxation energy initial prediction model using the second training data set to obtain the catalyst system relaxation energy prediction model;

[0031] FIG7 shows a schematic diagram of a catalyst system relaxation energy prediction model trained in an embodiment of the present application;

[0032] FIG8 is a schematic diagram showing a catalyst system relaxation energy prediction model for training multiple tasks in an embodiment of the present application;

[0033] FIG9 shows a schematic diagram of a catalyst system relaxation energy prediction model trained in an embodiment of the present application;

[0034] FIG10 is a schematic diagram showing a method for constructing a catalyst system relaxation energy prediction model that is pre-trained and not subjected to multi-task training in an embodiment of the present application;

[0035] FIG11 is a schematic diagram showing a method for constructing a relaxation energy prediction model for a catalyst system for multiple tasks in an embodiment of the present application;

[0036] FIG12 shows a comparison of the effects of the catalyst system relaxation energy prediction model and the control group in the embodiment of the present application;

[0037] FIG13 shows an exemplary structural block diagram of a device for constructing a catalyst system relaxation energy prediction model in one embodiment of the present application;

[0038] FIG14 shows a schematic structural diagram of a catalyst system relaxation energy prediction device according to one embodiment of the present application;

[0039] FIG15 illustrates an example system including an example computing device representative of one or more systems and / or devices that may implement the various methods described herein. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted.

[0041] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0042] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0043] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Thus, the first component discussed below can be referred to as the second component without departing from the teachings of the concepts of the present application. As used herein, the term "and / or" and similar terms include all combinations of any, multiple, and all of the associated listed items.

[0044] Before introducing the embodiments of the present application in detail, some of the terms involved in the embodiments of the present application are first explained to facilitate understanding by those skilled in the art.

[0045] Catalyst: A catalyst generally refers to a substance that increases the reaction rate without changing the overall standard Gibbs free energy of the reaction. Generally speaking, a catalyst is a substance that participates in the intermediate stages of a chemical reaction and selectively alters the reaction rate, while its quantity and chemical properties remain essentially unchanged before and after the reaction. Alternatively, it can be described as a substance that increases the reaction rate without altering the chemical equilibrium, and its mass and chemical properties remain unchanged before and after the reaction. According to statistics, catalysts are used in over 90% of industrial processes, such as those in the chemical, petrochemical, biochemical, and environmental protection sectors. Catalysts vary widely and can be categorized as liquid or solid based on their state. They can also be categorized as homogeneous or heterogeneous based on the phase of the reaction system. Homogeneous catalysts include acids, bases, soluble transition metal compounds, and peroxides. Catalysts play a vital role in the modern chemical industry. For example, iron catalysts are used in the production of synthetic ammonia, vanadium catalysts are used in the production of sulfuric acid, and various catalysts are employed in the polymerization of ethylene and the production of three major synthetic materials, rubber, from butadiene. Finding suitable catalysts for chemical reactions has always been an important topic in the chemical industry and an important means to control reaction rates and improve reaction safety.

[0046] Catalyst system: also known as adsorbate-catalyst system, refers to a system composed of multiple atoms or molecules that act as catalysts, and may also include atoms or molecules that act as reactants adsorbed on the catalyst. Because the catalyst system contains multiple atoms or molecules, and as the reactants that act as adsorbates continue to react, the forces between the atoms or molecules also change, and the energy of the system will also change until the reaction is completed and the system tends to a steady state. Therefore, for each microstructure of the catalyst system, it corresponds to a system energy, and at this time there will be a corresponding force relationship between the atoms or molecules therein. Usually, the difference between the system energy corresponding to the initial structure of a catalyst system and the system energy corresponding to the final structure after the reaction is completed is called the relaxation energy of the system, which is used to measure the performance of the catalyst. The greater the relaxation energy of the catalyst system, the better the performance of the catalyst.

[0047] Graph Neural Networks: Graph Neural Networks (GNNs) are a general term for algorithms that use neural networks to learn from graph-structured data, extracting and discovering features and patterns within it to meet the demands of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. By applying specific strategies to the nodes and edges within a graph, GNNs transform graph data into standardized representations, which are then fed into a variety of neural networks for training. They achieve excellent results in tasks such as node classification, edge information propagation, and graph clustering.

[0048] Topology: A topology is a simplified and regularized representation of the relationships between entities, displaying quantitative information. A topology conveys quantitative information graphically and is an effective way to represent entity relationships.

[0049] Fig. 1 shows a schematic diagram of a catalyst research and development process according to some embodiments of the present application. As shown in Figure 1, when developing a catalyst, a computing device 110 and an experimental device 120 are usually used to efficiently screen the catalyst. First, a large number of catalysts are predicted for relaxation energy using a computing device, and a catalyst with a good prediction result is selected. Then, the selected catalyst is experimentally verified using the experimental device 120, and the catalyst whose experimental performance meets the expected result is screened out. This catalyst screening method with the help of a computing device can greatly improve the efficiency of catalyst research and development, but higher requirements are placed on the accuracy and speed of the computing device 110 prediction of relaxation energy. For this reason, it is necessary to construct a catalyst system relaxation energy prediction model. However, due to the complex atomic structure and rapid structural changes during chemical reactions of the catalyst system, its relaxation energy prediction model is difficult to establish and has poor accuracy. In the related art scheme, a model can be constructed by a quantum mechanics method, but this method has many shortcomings. The quantum mechanics model usually uses a density functional theory (Density Functional Theory) method based on density functional theory (Density Functional) to realize molecular simulation (Molecular Simulation), and then predicts the relaxation energy of the catalyst system based on molecular simulation. Since the construction and operation of this model require calculations on each molecule or atom, the computational complexity is very large and the response speed is slow, whether it is modeling or running the model, which makes it difficult to meet the needs of actual applications.

[0050] To this end, the present application proposes a catalyst system relaxation energy prediction model construction method for constructing a catalyst system relaxation energy prediction model. The model can be run on a computing device 110 to achieve accurate prediction of the catalyst system relaxation energy.

[0051] FIG2 illustrates an exemplary application scenario 200 of a catalyst system relaxation energy prediction model construction method according to some embodiments of the present application. Application scenario 200 may include a server 210, a terminal device 220, and a server 230. Server 210, terminal device 220, and server 230 are communicatively coupled together via a network 240. Network 240 may be, for example, a wide area network (WAN), a local area network (LAN), a wireless network, a public telephone network, an intranet, or any other type of network known to those skilled in the art.

[0052] In some embodiments, the catalyst system relaxation energy prediction model construction method can be mainly run on the server 210. On the server 210, a first training data set is obtained, the first training data set includes a plurality of first training data, each first training data includes a first training sample and a corresponding first sample label, wherein the first training sample includes the first catalyst system structure information, and the first sample label includes the system energy information corresponding to the first catalyst system structure information. In some embodiments, the first training data set can be stored on the server 210, or can be obtained from other servers or terminals through the network 240. Then, on the server 210, the catalyst system energy prediction model is trained using the first training data set to obtain a pre-trained catalyst system energy prediction model, and based on the pre-trained catalyst system energy prediction model, an initial catalyst system relaxation energy prediction model is constructed. Then, on the server 210, a second training data set is obtained, the second training data set includes a plurality of second training data, each second training data includes a second training sample and a corresponding second sample label, wherein the second training sample includes the second catalyst system structure information, and the second sample label includes the relaxation energy information corresponding to the second catalyst system structure information. In some embodiments, the second training data set may be stored on the server 210 or obtained from another server or terminal via the network 240. Finally, on the server 210, the catalyst system relaxation energy initial prediction model is trained using the second training data set to obtain the catalyst system relaxation energy prediction model.

[0053] In some embodiments, the catalyst system relaxation energy prediction model construction method can also be mainly run on the terminal device 220 or the server 230. It should be noted that the server 210, the terminal device 220 and the server 230 can all include a medium and / or device capable of persistently storing information, and / or a tangible storage device. Therefore, a computer-readable storage medium refers to a non-signal-bearing medium. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented in a method or technology suitable for storing information (such as computer-readable instructions, data structures, program modules, logical elements / circuits or other data). As understood by those of ordinary skill in the art, an instance of the server 210 can be an independent physical server, or a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDNs, and big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected by wired or wireless communication, and this application is not limited thereto. The server 210 may present the data allocation strategy to be determined to the developer through the terminal device 220 and interact with the developer to achieve visual determination of the development strategy.

[0054] The terminal device 220 can be any type of mobile computing device, including a mobile computer (e.g., a personal digital assistant (PDA), laptop computer, notebook computer, tablet computer, netbook computer, etc.), a mobile phone (e.g., a cellular phone, smartphone, etc.), a wearable computing device (e.g., a smartwatch, a head-mounted device, including smart glasses, etc.), or other types of mobile devices. In some embodiments, the terminal device 220 and the server 230 can also be stationary computing devices, such as a desktop computer, a game console, a smart TV, etc. In addition, when the application scenario 200 includes multiple terminal devices 220, the multiple terminal devices 220 can be computing devices of the same or different types.

[0055] As shown in Figure 2, the terminal device 220 may include a display screen and a terminal application that can interact with the terminal user via the display screen. The terminal application can be a local application, a web application, or a mini-program (LiteApp, such as a mobile app, WeChat app) as a lightweight application. In the case where the terminal application is a local application that needs to be installed, the terminal application can be installed in the terminal device 220. In the case where the terminal application is a web application, the terminal application can be accessed through a browser. In the case where the terminal application is a mini-program, the terminal application can be opened directly on the terminal device 220 by searching for relevant information of the terminal application (such as the name of the terminal application, etc.), scanning a graphic code of the terminal application (such as a barcode, a QR code, etc.), etc., without installing the terminal application.

[0056] In some embodiments, the aforementioned application scenario 200 may be a distributed system comprised of a server 230, which may, for example, constitute a blockchain system. Blockchain is a novel application model for computer technologies, including distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (for anti-counterfeiting) and generate the next block. Blockchain may include an underlying blockchain platform, a platform product and service layer, and an application service layer.

[0057] FIG3 shows an exemplary flow chart of a method 400 for constructing a catalyst system relaxation energy prediction model according to some embodiments of the present application. As shown in FIG3 , the method 300 includes steps S310 , S320 , S330 , S340 , and S350 .

[0058] In step S310, a first training data set is obtained, where the first training data set includes a plurality of first training data, each first training data includes a first training sample and a corresponding first sample label, wherein the first training sample includes first catalyst system structure information, and the first sample label includes system energy information corresponding to the first catalyst system structure information.

[0059] The first catalyst system structure information may be the structure information of the catalyst system represented by the first training sample, and may represent the structure of the catalyst system. The system energy information corresponding to the first catalyst system structure information may represent the system energy of the catalyst system represented by the first catalyst system structure information.

[0060] In some embodiments, the system energy information corresponding to the first catalyst system structure information can be obtained through measurement or quantum mechanical calculation and stored for easy retrieval. In other embodiments, the first sample tag can also include atomic force information corresponding to the first catalyst system structure information. The atomic force information can be obtained through measurement or quantum mechanical calculation and stored for easy retrieval.

[0061] In step S320 , the catalyst system energy prediction model is trained using the first training data set to obtain a pre-trained catalyst system energy prediction model.

[0062] In some embodiments, the catalyst system energy prediction model may include a graph neural network, and training the catalyst system energy prediction model using a first training data set includes training the parameters of the graph neural network using the first training data set to obtain a graph neural network that can predict the catalyst system energy based on the catalyst system structure.

[0063] In step S330 , an initial catalyst system relaxation energy prediction model is constructed based on the pre-trained catalyst system energy prediction model.

[0064] In some embodiments, a pre-trained catalyst system energy prediction model can be used as the initial prediction model for the catalyst system's relaxed energy. Because the pre-trained catalyst system energy prediction model is a multi-task model (the model simultaneously predicts the system energy and system forces of the catalyst system based on the catalyst system structure), its parameter tuning process is more rigorous than the training of a single-task model (for example, it can be more comprehensive when calculating losses), making the model's parameter tuning more efficient, converge faster, and the resulting parameters more adaptable.

[0065] The output of the catalyst system relaxation energy initial prediction model includes outputs similar to those of the catalyst system relaxation energy initial prediction model (for example, both include one-dimensional output and matrix output). Therefore, using the pre-trained catalyst system energy prediction model as the catalyst system relaxation energy initial prediction model can determine a better initial parameter value for the subsequent construction of the model.

[0066] In step S340, a second training data set is obtained, where the second training data set includes a plurality of second training data, each second training data includes a second training sample and a corresponding second sample label, wherein the second training sample includes second catalyst system structure information, and the second sample label includes relaxation energy information corresponding to the second catalyst system structure information.

[0067] In some embodiments, the second catalyst system structure information may be structure information of the catalyst system represented by the second training sample, and may characterize the structure of the catalyst system. The relaxation energy information corresponding to the second catalyst system structure information may represent the relaxation energy of the catalyst system represented by the second catalyst system structure information.

[0068] In some embodiments, the relaxation energy corresponding to the second catalyst system structure information can be obtained by measurement or quantum mechanical calculation and stored for retrieval upon use. The second sample label can also include atomic displacement information corresponding to the second catalyst system structure information. The atomic displacement information corresponding to the second catalyst system structure information can be obtained by measurement or quantum mechanical calculation and stored for retrieval upon use.

[0069] In step S350 , the catalyst system relaxation energy initial prediction model is trained using the second training data set to obtain a catalyst system relaxation energy prediction model.

[0070] In some embodiments, the initial prediction model of the catalyst system relaxation energy may include a graph neural network, and training the initial prediction model of the catalyst system relaxation energy using a second training data set includes further training on the parameters pre-trained by the graph neural network using the second training data set to obtain a graph neural network that can predict the catalyst system relaxation energy based on the catalyst system structure.

[0071] Therefore, method 400 constructs an initial prediction model of the catalyst system relaxation energy based on the pre-trained catalyst system energy prediction model, which enables the catalyst system relaxation energy initial prediction model to be quickly determined. Then, the catalyst system relaxation energy initial prediction model is adjusted by the second training data set to obtain the catalyst system relaxation energy prediction model, which enables a catalyst system relaxation energy prediction model with high precision and fast response to be quickly obtained. It can be seen that, on the one hand, the catalyst system relaxation energy prediction model constructed by method 400 significantly improves the accuracy of relaxation energy prediction due to the adoption of a two-stage training process, thereby avoiding the problem of low calculation accuracy in related technologies; on the other hand, the catalyst system relaxation energy prediction model constructed according to some embodiments of the present application can directly predict relaxation energy based on the initial structure information of the system, without the need for iterative calculation, and with a faster calculation speed, thereby avoiding the problems of high computational complexity, cumbersome calculation process, and slow calculation speed in traditional quantum mechanics methods and related technologies. Therefore, by method 400, a high-quality catalyst system relaxation energy prediction model can be quickly and easily constructed.

[0072] In some embodiments, the first sample tag may further include atomic force information corresponding to the first catalyst system structure information, and the second sample tag may further include atomic displacement information corresponding to the second catalyst system structure information.

[0073] Among them, the atomic force information can represent the information of the atomic forces in the catalyst system having the system structure represented by the first catalyst system structure information, and can include the force magnitude and force direction. The atomic displacement information represents the information of the atomic displacement in the catalyst system. The atomic displacement of an atom in the catalyst system can be obtained by subtracting the initial position of the atom from the final position of the atom (i.e., the position of the atom when the system is in a relaxed state). The atomic displacement information can include the atomic displacement magnitude and displacement direction.

[0074] In some embodiments, because the second training data contains a variety of expected prediction data (e.g., relaxation energies and atomic displacements), the model trained using the second dataset can handle multiple tasks simultaneously. Furthermore, because the initial catalyst system relaxation energy prediction model is a multi-task model, adjusting the model parameters using the second training data allows for rapid convergence and high model prediction accuracy.

[0075] In the present embodiment, the catalyst system relaxation energy initial prediction model for multitasking is constructed according to the catalyst system energy prediction model for multitasking of pre-training, which makes it possible to quickly determine the catalyst system relaxation energy initial prediction model for multitasking.Then, the catalyst system relaxation energy initial prediction model for multitasking is adjusted by the second training data set, to obtain the catalyst system relaxation energy prediction model for multitasking. Due to the training of the first and second phases, model training is used to adapt to multitasking (for example, the first phase includes prediction system energy and atomic force two tasks, and the second phase includes prediction relaxation energy and atomic displacement two tasks), and the first and second phases are corresponding to the dimension of the expected output of model (for example, the data dimension of system energy and relaxation energy is identical, and the data dimension of atomic force and atomic displacement is identical), therefore the training process of the second phase can be faster and more efficient, and the prediction accuracy of final catalyst system relaxation energy prediction model is also higher. It can be seen that, on the one hand, the catalyst system relaxation energy prediction model significantly improves the accuracy of relaxation energy prediction due to the two-stage training process, thus avoiding the problem of low calculation accuracy of related technologies; on the other hand, the catalyst system relaxation energy prediction model can directly predict the relaxation energy based on the initial structure information of the system, without the need for iterative calculation, and the calculation speed is faster, thus avoiding the problems of high computational complexity, cumbersome calculation process, and slow calculation speed of traditional quantum mechanics methods and related technologies; on the other hand, the catalyst system relaxation energy prediction model is trained for multiple tasks in the first and second stages, so its training efficiency is higher and the obtained model prediction accuracy is better.

[0076] Figure 4 shows a schematic flow chart of training a catalyst system energy prediction model using a first training data set to obtain a pre-trained catalyst system energy prediction model. As shown in Figure 4, in some embodiments, step S320 includes steps S410, S420, S430, and S440.

[0077] In step S410 , for each first training data in the first training data set, a first training sample of the first training data is input into a catalyst system energy prediction model to obtain a first output result corresponding to the first training data.

[0078] In step S420, for each first training data point in the first training data set, a first loss corresponding to the first training data point is calculated based on the first output result corresponding to the first training data point and the first sample label of the first training data point. In some embodiments, step S420 includes calculating the first loss corresponding to each first training data point in the first training data set based on the catalyst system energy prediction result corresponding to the first training data point and the system energy information of the first training data point.

[0079] In step S430, a first target loss of the catalyst system energy prediction model is determined based on the first loss corresponding to each first training data in the first training data set. In some embodiments, the first target loss can be calculated using a loss function.

[0080] In some embodiments, when the first sample label further includes: atomic force information corresponding to the first catalyst system structure information, the first output result includes a catalyst system energy prediction result and an atomic force information prediction result; step S430 includes: for each first training data in the first training data set, based on the atomic force information prediction result corresponding to the first training data and the atomic force information of the first training data, calculating the second loss corresponding to the first training data; and determining the first target loss of the catalyst system energy prediction model based on the first loss and second loss corresponding to each first training data in the first training data set. When calculating the first target loss, all calculated first losses and all calculated second losses may be summed, and the summed result may be used as the first target loss.

[0081] In step S440, based on the first target loss, the parameters of the catalyst system energy prediction model are iteratively updated until the first target loss satisfies a first preset condition, thereby obtaining a pre-trained catalyst system energy prediction model. Therefore, the pre-trained catalyst system energy prediction model can accurately predict catalyst system energy based on the structure of the catalyst system.

[0082] In some embodiments, the catalyst system energy prediction model includes a catalyst system energy-force prediction model, and the first sample label further includes first atomic force information corresponding to the first catalyst system structure information. In some embodiments, the first atomic force information corresponding to the first catalyst system structure information can be obtained through experimental measurement or quantum mechanical calculation and stored for retrieval when needed. The first output result includes the catalyst system energy prediction result and the atomic force information prediction result. In some embodiments, the catalyst system energy prediction result can be expressed as a scalar, and the atomic force information prediction result can be expressed as a matrix.

[0083] In some embodiments, the first preset condition includes at least one of the following: the current first target loss is less than a predetermined threshold; the number of iterations corresponding to the current first target loss reaches a predetermined number. For example, the predetermined number is set to 500, and when the number of iterations exceeds 500, the iteration is stopped.

[0084] In some embodiments, step S420 includes the following steps.

[0085] First, for each first training data point in the first training data set, a first sub-loss is calculated based on the catalyst system energy prediction result corresponding to the first training data point and the system energy information of the first training data point. In some embodiments, the first sub-loss can be obtained by calculating the difference between the catalyst system energy prediction result corresponding to the first training data point and the system energy information of the first training data point.

[0086] Then, for each first training data point in the first training data set, a second sub-loss corresponding to the first training data point is calculated based on the atomic force information prediction result corresponding to the first training data point and the atomic force information of the first training data point. In some embodiments, the second sub-loss can be obtained by calculating the difference in each dimension between the matrix representing the atomic force information prediction result and the matrix representing the atomic force information, and then combining them into a scalar.

[0087] Finally, for each first training data in the first training data set, a first loss is determined based on the first sub-loss and the second sub-loss corresponding to the first training data. In some embodiments, determining the first loss based on the first sub-loss and the second sub-loss may include calculating the sum of the first sub-loss and the second sub-loss to obtain.

[0088] In some embodiments, the atomic force information of the catalyst system can be represented as a three-dimensional matrix, where the data in each dimension represent the forces acting on the atoms in the x, y, and z directions. Furthermore, the system structure information of the catalyst system can also be represented as a three-dimensional matrix, where the data in each dimension represent the three-dimensional coordinates of the atoms. This close dimensionality between the input and output facilitates parameter training of the catalyst system energy-force prediction model.

[0089] Figure 5 shows a schematic diagram of a training catalyst system energy prediction model according to some embodiments of the present application. As shown in Figure 5, first obtain the 3D structure of the catalyst system (including the coordinates and serial numbers of each atom), and the system energy and atomic force corresponding to the 3D structure of the catalyst system. Exemplarily, the system energy and atomic force corresponding to the 3D structure of the catalyst system can be calculated using quantum mechanics. Then, a topological map of the catalyst system is constructed based on the distance between atoms. Then, with the topological map of the catalyst system structure as input, the system energy and atomic force corresponding to the catalyst system structure are used as expected outputs, and the parameters of the graph neural network are trained to obtain a catalyst system energy prediction model.

[0090] Figure 6 shows a schematic flow diagram of training the catalyst system relaxation energy initial prediction model using the second training data set to obtain the catalyst system relaxation energy prediction model. As shown in Figure 6, in some embodiments, step S350 includes steps S610, S620, S630, and S640.

[0091] In step S610 , for each second training data in the second training data set, the second catalyst system structure information of the second training data is input into the catalyst system relaxation energy initial prediction model to obtain a second output result corresponding to the second training data.

[0092] In some embodiments, the second output includes a catalyst system relaxation energy prediction result and an atomic displacement information prediction result. In some embodiments, the catalyst system relaxation energy prediction result is represented by a scalar, and the atomic displacement information prediction result is represented by a matrix.

[0093] In step S620 , for each second training data in the second training data set, a third loss corresponding to the second training data is calculated based on the second output result and the second sample label corresponding to the second training data.

[0094] In some embodiments, the second output result includes a catalyst system relaxation energy prediction result; step S620 includes: for each second training data in the second training data set, based on the catalyst system relaxation energy prediction result corresponding to the second training data and the relaxation energy information of the second training data, calculating the third loss corresponding to the second training data

[0095] In some embodiments, the third loss may be obtained by calculating the difference between the catalyst system relaxation energy prediction result and the relaxation energy information of the second training data.

[0096] In step S630, a second target loss of the catalyst system relaxation energy initial prediction model is determined based on the third loss corresponding to each second training data in the second training data set. In some embodiments, the second target loss can be obtained using a loss function.

[0097] In some embodiments, the second sample label also includes: atomic displacement information corresponding to the second catalyst system structure information; the second output result also includes an atomic displacement information prediction result; step S630 includes: for each second training data in the second training data set, based on the atomic displacement information prediction result corresponding to the second training data and the atomic displacement information of the second training data, calculating the fourth loss corresponding to the second training data; according to the third loss and the fourth loss corresponding to each second training data in the second training data set, determining the second target loss of the initial prediction model of the catalyst system relaxation energy.

[0098] In some embodiments, the fourth loss can be obtained by calculating the difference in each dimension between matrices and then combining them into a scalar. In some embodiments, for each element in the second training dataset, the third loss and the fourth loss are summed to obtain a comprehensive loss. The comprehensive loss corresponding to each element in the second training dataset is then summed, and the summed result is determined as the second target loss.

[0099] In step S640 , based on the second target loss, the parameters of the initial prediction model of the catalyst system relaxation energy are iteratively updated until the second target loss satisfies a second preset condition, so as to obtain a catalyst system energy prediction model.

[0100] In some embodiments, the initial prediction model of catalyst system relaxation energy may include a graph neural network, and training the initial prediction model of catalyst system relaxation energy using a second training data set includes further training on the parameters pre-trained by the graph neural network using the second training data set to obtain a graph neural network that can simultaneously predict catalyst system relaxation energy and atomic displacement based on the catalyst system structure.

[0101] In some embodiments, the second preset condition includes at least one of the following: the current second target loss is less than a predetermined threshold; or the number of iterations corresponding to the current second target loss reaches a predetermined number. The number of iterative updates of the parameters of the initial catalyst system relaxation energy prediction model may be less than the number of iterative updates of the parameters of the catalyst system energy prediction model.

[0102] In some embodiments, the initial prediction model of the catalyst system relaxation energy may include a graph neural network, and training the initial prediction model of the catalyst system relaxation energy using a second training data set includes further training on the parameters pre-trained by the graph neural network using the second training data set to obtain a graph neural network that can predict the catalyst system relaxation energy based on the catalyst system structure.

[0103] In some embodiments, the first output result includes a catalyst system energy prediction result and an atomic force information prediction result; the catalyst system energy prediction model includes a topology map determination submodel and a graph neural network submodel. Step S410 includes the following steps. First, for each first training data in the first training data set, the first catalyst system structure information corresponding to the first training data is input into the topology map determination submodel to obtain a topology map corresponding to the first catalyst system structure information, and the topology map is used to characterize the topological association between each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data. Then, for each first training data in the first training data set, the topology map corresponding to the first catalyst structure information corresponding to the first training data is input into the graph neural network submodel to obtain the catalyst system energy prediction result and the atomic force information prediction result. In some embodiments, the graph neural network submodel includes a graph neural network, and the parameters of the graph neural network are obtained through training.

[0104] In some embodiments, the topological graph comprises a plurality of nodes and edges connecting the plurality of nodes in pairs, each node represents an atom in the corresponding catalyst system, and each edge represents the association between the two atoms corresponding to the edge. For each first training data in the first training data set, the first catalyst system structure information corresponding to the first training data is input into the topological graph determination submodel to obtain the topological graph corresponding to the first catalyst system structure information, which comprises the following steps. First, for each first training data in the first training data set, the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data are determined based on the first catalyst system structure information corresponding to the first training data. In some embodiments, the atoms may include adsorbate atoms and catalyst atoms, etc. Then, for each first training data in the first training data set, the positions of each node in the topological graph corresponding to the first catalyst system structure information and the distances between each pair of atoms are determined based on the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data. For each two-node combination in the topological graph, in response to the distance between the atoms corresponding to the two nodes in the combination being less than or equal to the predetermined distance threshold, the value of the edge connecting the two nodes is 1; for each two-node combination in the topological graph, in response to the distance between the atoms corresponding to the two nodes in the combination being greater than the predetermined distance threshold, the value of the edge connecting the two nodes is 0.

[0105] In some embodiments, obtaining a first training data set includes the following steps: obtaining three-dimensional structural information of each sample catalyst system for a plurality of sample catalyst systems, the three-dimensional structural information including the three-dimensional coordinates of each atom in the corresponding sample catalyst system; determining system energy information and atomic force information of each sample catalyst system using quantum mechanics methods based on the three-dimensional structural information of each sample catalyst system in the plurality of sample catalyst systems; and constructing first training data based on the three-dimensional structural information of each sample catalyst system and its corresponding system energy information and atomic force information. In some embodiments, the system energy information and atomic force information of the catalyst system can be obtained using quantum mechanics calculations based on the distances between atoms in the catalyst system structure using quantum mechanics methods.

[0106] Fig. 7 shows the schematic diagram of the training catalyst system relaxation energy prediction model according to some embodiments of the present application. As shown in Figure 7, first obtain the 3D structure of the catalyst system (comprising the coordinates and serial numbers of each atom), and the relaxation energy corresponding to the catalyst system 3D structure. Exemplarily, the relaxation energy corresponding to the catalyst system 3D structure is obtained by calculating the difference between the system energy of the initial 3D structure of the catalyst system and the system energy of the final 3D structure. Then, a topological map of the catalyst system is constructed based on the distance between atoms. Then, with the topological map of the catalyst system structure as input, the relaxation energy corresponding to the catalyst system structure is used as expected output, and the parameters of the graph neural network are trained (for example, parameters are fine-tuned based on the initial parameters), and catalyst relaxation energy prediction model is obtained. In certain embodiments, the initial parameters of the graph neural network can use the parameters of the catalyst system energy prediction model.

[0107] Fig. 8 shows a schematic diagram of a catalyst system relaxation energy prediction model for multi-tasks according to the training of some embodiments of the present application. As shown in Figure 8, first obtain the 3D structure of the catalyst system (comprising the coordinates and serial numbers of each atom), and the relaxation energy and atomic displacement corresponding to the 3D structure of the catalyst system. Exemplarily, the relaxation energy corresponding to the 3D structure of the catalyst system is obtained by calculating the difference between the system energy of the initial 3D structure of the catalyst system and the system energy of the final 3D structure. The atomic displacement corresponding to the 3D structure of the catalyst system is obtained by calculating the difference between the coordinates of the individual atoms of the initial 3D structure of the catalyst system and the coordinates of the individual atoms of the final 3D structure. Figure 9 shows a schematic diagram of the atomic displacement according to some embodiments of the present application. As shown in Figure 9, the atomic displacement of a certain atom in the catalyst system can be obtained by subtracting the initial position of the atom from the final position of the atom (i.e., the position of the atom when the system is in a relaxed state) (i.e., the vector R in Figure 9).

[0108] In some embodiments, the atomic coordinate information of the catalyst system can be represented in the form of a three-dimensional matrix, and the data on the three dimensions respectively represent the position mapping of the atoms in the x, y, and z directions. In addition, the atomic displacement of the catalyst system can also be represented in the form of a three-dimensional matrix, and the data on the three dimensions respectively represent the position changes of the atoms in the x, y, and z directions. In combination with the pre-trained embodiments in the foregoing, it can be found that the dimension of the atomic displacement is the same as the dimension of the atomic force, and the dimension of the relaxation energy is the same as the dimension of the system energy, which is conducive to quickly setting appropriate parameters during training. In some embodiments, the initial parameters of the graph neural network can use the parameters of the catalyst system energy prediction model (i.e., the parameters obtained by pre-training in the foregoing embodiments).

[0109] In some embodiments, step S320 includes the following steps: based on a preset first hyperparameter set, the catalyst system energy prediction model is trained using a first training data set to obtain a pre-trained catalyst system energy prediction model, wherein the preset first hyperparameter set includes a preset first learning rate. And, step S350 includes the following steps: based on a preset second hyperparameter set, the catalyst system relaxation energy initial prediction model is trained using a second training data set to obtain a catalyst system relaxation energy prediction model, wherein the second hyperparameter set includes a preset second learning rate, and the second learning rate is less than the first learning rate. In some embodiments, the second learning rate is equal to or less than one tenth of the first learning rate. Therefore, the learning rate when training the catalyst system energy prediction model using the first training data set is much greater than the learning rate when training the catalyst system relaxation energy initial prediction model using the second training data set, because the initial parameters of the latter have been determined by the former, and therefore only fine-tuning on the initial parameters is required. It can be seen that the model construction method of this embodiment can more quickly and accurately construct a catalyst system relaxation energy prediction model.

[0110] In some embodiments, the first sample label further includes: atomic force information corresponding to the first catalyst system structure information, and the second sample label further includes: atomic displacement information corresponding to the second catalyst system structure information. Since the models trained in the pre-training and fine-tuning stages are both multi-task prediction models, and their output dimensions are mutually corresponding (the data dimensions of system energy and relaxation energy are the same, and the data dimensions of atomic forces and atomic displacements are the same), the initial parameters are determined through pre-training, and then fine-tuned in the fine-tuning stage, so that appropriate model parameters can be quickly determined.

[0111] This application also discloses a method for predicting the relaxation energy of a catalyst system, comprising: obtaining system structure information of a catalyst system to be predicted; and inputting the system structure information of the catalyst system to be predicted into a catalyst system relaxation energy prediction model disclosed in any embodiment herein to obtain the relaxation energy of the catalyst system. In some embodiments, the catalyst system relaxation energy prediction model can be run on a computing device 110 as shown in FIG1 .

[0112] Figure 10 shows a schematic diagram of a method for constructing a catalyst system relaxation energy prediction model without multi-task training according to some embodiments of the present application. As shown in Figure 10, the pre-training stage is first completed using the first training data set. In the pre-training stage, the first training data set is first obtained, which contains 2 million data, each of which contains the structure of the catalyst system and the corresponding system energy information. Then, the corresponding topological map is determined according to the catalyst system structure. Taking the topological map of the catalyst system structure as input and the system energy corresponding to the catalyst system structure as the expected output, the parameters of the graph neural network are trained to obtain the first parameters of the graph neural network as the pre-training result.

[0113] Then, the fine-tuning phase is completed using the second training data set. The second training data set contains 460,000 data, each of which contains the initial structure of the catalyst system and the corresponding relaxation energy. Then, the corresponding topology map is determined based on the initial structure of the catalyst system. The graph neural network is trained using the topology map of the initial structure of the catalyst system as input and the relaxation energy corresponding to the initial structure of the catalyst system as the expected output. The training process is as follows: the parameters of the graph neural network are adjusted from the first parameter until the actual output of the graph neural network is close to the expected output, and the parameters of the graph neural network at this time are determined as the second parameter as the training result. Finally, the construction of the catalyst system relaxation energy prediction model is completed. The catalyst system relaxation energy prediction model includes a topology map generator and a graph neural network. The topology map generator generates a corresponding topology map based on the 3D structure of the catalyst system. The graph neural network predicts the relaxation energy of the catalyst system based on the topology map of the catalyst system. The parameters of the graph neural network are the second parameters.

[0114] Figure 11 shows a schematic diagram of a method for constructing a catalyst system relaxation energy prediction model for multiple tasks according to some embodiments of the present application. First, the pre-training stage is completed using the first training data set. In the pre-training stage, the first training data set is first obtained, which contains 2 million data, each of which contains the structure of the catalyst system and the corresponding system energy and atomic force information. Then, the corresponding topological map is determined according to the catalyst system structure. Taking the topological map of the catalyst system structure as input and the system energy and atomic force information corresponding to the catalyst system structure as the expected output, the parameters of the graph neural network are trained to obtain the first parameters of the graph neural network as the pre-training result. It can be seen that in the pre-training stage, although the input of the graph neural network is only a topological map representing the system structure, its output is the system energy and atomic force, that is, the graph neural network is trained to learn two tasks, so it belongs to multi-task learning. Multi-task learning is conducive to establishing a more rigorous and complete supervision function, making the process of training the graph neural network more efficient and faster, and the obtained parameters are also more adaptable.

[0115] The fine-tuning phase is then completed using a second training dataset. The second training dataset contains 460,000 data points, each of which contains the initial structure of the catalyst system and the corresponding relaxation energy and atomic displacements. A corresponding topological map is then determined based on the initial structure of the catalyst system. The graph neural network is trained using the topological map of the initial structure as input and the relaxation energy and atomic displacements corresponding to the initial structure as the expected output. The training process is as follows: the parameters of the graph neural network are adjusted starting from the first parameter until the actual output of the graph neural network approaches the expected output. The parameters of the graph neural network at this point are determined as the second parameter and used as the training result. Finally, the catalyst system relaxation energy prediction model is constructed. The catalyst system relaxation energy prediction model comprises a topological map generator and a graph neural network. The topological map generator generates a corresponding topological map based on the 3D structure of the catalyst system. The graph neural network predicts the relaxation energy and atomic displacements of the catalyst system based on the topological map of the catalyst system. The parameters of the graph neural network are the second parameter. As can be seen, during the fine-tuning phase, although the input of the graph neural network is only the topological map representing the system structure, its output is relaxation energy and atomic displacements. This means that the graph neural network is trained to learn two tasks, thus still belonging to multi-task learning. Furthermore, the learning of the two tasks in the fine-tuning phase corresponds to the learning of the two tasks in the pre-training phase, because their output data dimensions are the same. This allows the GNN parameters obtained in the pre-training phase to be used as the initial parameters of the GNN in the fine-tuning phase, allowing the most appropriate GNN parameters to be quickly determined during the fine-tuning phase. This results in a higher accuracy of the GNN in predicting system relaxation energies and atomic displacements after fine-tuning.

[0116] Table I Examples of hyperparameters of graph neural networks in the pre-training phase

[0117] Table I shows the hyperparameters for training the graph neural network during the pre-training phase in the embodiment shown in Figure 10. As shown in Table I, the graph neural network has 16 layers, a learning rate of 0.0004, a batch size of 384, and 20 epochs. These hyperparameters define the pre-training process. An epoch is the process of training all samples in the training dataset once. During an epoch, the training algorithm feeds all samples into the model in a pre-set order for forward propagation, loss calculation, backpropagation, and parameter updates. A batch refers to a group of samples fed into the model at once. During neural network training, the training data is often very large, such as tens of thousands or even hundreds of thousands of samples. If all these tens of thousands of data points were fed into the model at once, the requirements on computer performance and the learning ability of the neural network model would be too high. Instead, the training data can be divided into multiple batches, and the samples from each batch are then fed into the model in batches for forward propagation, loss calculation, backpropagation, and parameter updates. The "batch size" refers to the number of samples in each batch. The learning rate, also known as the learning rate, represents the rate at which information accumulates over time in a neural network. It's one of the most influential hyperparameters, and if we can only adjust one, it's the best choice. Compared to other hyperparameters, the learning rate controls the effective capacity of the model in a more complex way. When the learning rate is optimal, the model's effective capacity is maximized. Therefore, one of the key hyperparameters to set for neural network training is the learning rate.

[0118] Table II Examples of hyperparameters of graph neural networks in the fine-tuning stage

[0119] Table II shows the hyperparameters for training the graph neural network in the fine-tuning phase of the embodiment shown in Figure 11. As shown in Table II, the number of layers (Layer Number) of the graph neural network is 16, the learning rate (Learning rate) is 0.00004, the batch size (Batch size) is 384, the rounds (Epochs) is 20, the relaxation energy loss factor (Relexed Energy Loss Factor) is 1, and the atomic displacement loss factor (Delta Postion Loss Factor) is 5. These hyperparameters limit the training process. Comparing Table II and Table I, it can be found that in the fine-tuning phase, the learning rate of training the graph neural network is one tenth of that in the pre-training phase. This is because in the fine-tuning phase, the initial parameters of the graph neural network are the training results of the pre-training phase, which are not far from the target results, so only fine-tuning is required. Therefore, the learning rate in the fine-tuning phase can be set much smaller than that in the pre-training phase. It can be seen that in the fine-tuning stage, the output head used to predict the system energy in the "atom-force" model is used to predict the system relaxation energy, and the output head used to predict the atomic force of the system in the "atom-force" model is also used to predict the atomic displacement of the system. Predicting the relaxation energy of the system is a graph-level task, and predicting the atomic displacement in the system is a node-level task. The two can be trained at the same time. This process is multi-task training. Predicting the position offset of the system is actually an auxiliary task. Its addition can further improve the model's prediction accuracy for the system relaxation energy. It can be seen that by constructing the model through the steps of this embodiment, the initial parameters of the model can be quickly determined first, and then fine-tuned based on the initial parameters to achieve efficient and accurate model construction.

[0120] Table III Hyperparameters of training graph neural networks in the pre-training phase

[0121] Table III shows the hyperparameters for training the graph neural network during the pre-training phase in the embodiment shown in FIG8 . As shown in Table III, the number of layers in the graph neural network is 16, the learning rate is 0.0004, the batch size is 384, and the number of epochs is 20. These hyperparameters define the pre-training process. As mentioned earlier, during neural network training, we often need to divide the training data into multiple batches; the specific number of samples in each batch is determined by the batch size. The learning rate, also known as the learning rate, represents the rate at which information accumulates over time in a neural network. The learning rate is one of the hyperparameters that most affects performance, and if we can only adjust one hyperparameter, this is the best choice. Compared to other hyperparameters, the learning rate controls the effective capacity of the model in a more complex way. When the learning rate is optimal, the effective capacity of the model is maximized. Therefore, to train a neural network, one of the key hyperparameters that needs to be set is the learning rate.

[0122] Table IV Hyperparameters of training graph neural networks in the fine-tuning phase

[0123] Table IV shows the hyperparameters for training the graph neural network in the fine-tuning phase of the embodiment shown in Figure 7. As shown in Table IV, the number of layers (Layer Number) of the graph neural network is 16, the learning rate (Learning rate) is 0.00004, the batch size (Batch size) is 384, and the rounds (Epochs) is 20. These hyperparameters limit the training process. Comparing Table IV and Table III, it can be found that in the fine-tuning phase, the learning rate of training the graph neural network is one tenth of that in the pre-training phase. This is because in the fine-tuning phase, the initial parameters of the graph neural network are the training results of the pre-training phase, which are not far from the target results, so only fine-tuning is required. Therefore, the learning rate in the fine-tuning phase can be set much smaller than that in the pre-training phase. It can be seen that by constructing a model through the steps of this embodiment, the initial parameters of the model can be quickly determined first, and then fine-tuned based on the initial parameters to achieve efficient and accurate model construction.

[0124] Figure 12 shows a rendering of a catalyst system relaxation energy prediction model according to some embodiments of the present application. In Figure 12, the control group selected a catalyst system relaxation energy prediction model that was directly constructed without a pre-training stage, and the embodiment of the present application adopted a catalyst system relaxation energy prediction model that was constructed after pre-training but without multi-task training as shown in Figure 10, and a catalyst system relaxation energy prediction model that was constructed after multi-task training as shown in Figure 11. The test set selected 4 authoritative test sets in the industry, namely ID (In Domain), OOD-Ads (Out Of Domain Adsorbate), OOD-Cat (Out Of Domain Catalyst) and OOD-Both (Out Of Domain Both) test sets, and their data volume is shown in Table V.

[0125] Table V Authoritative Test Set Examples

[0126] Using the four authoritative test sets shown in Table V, the catalyst relaxation energy prediction model constructed for the control group and the embodiments of the present application was tested, and the test results are shown in Figure 12. The prediction accuracy is measured by the prediction error, and the prediction error is measured using the MAE (Mean Absolute Error) indicator. The lower the value of this indicator, the higher the prediction accuracy of the model. In Figure 12, the black bar represents the control group, the grid-shaped bar represents the embodiment of the present application that has not been trained with too many tasks (referred to as the present application scheme 1), and the white bar represents the embodiment of the present application that has been trained with multiple tasks (referred to as the present application scheme 2). It can be clearly seen that on the four test sets, the prediction accuracy of the present application scheme 1 and the present application scheme 2 are both higher than that of the control group. In addition, the figure shows the average value (Average) of the four groups. It can be seen from the average value that the average prediction accuracy of the model of the present application scheme 1 is improved by 21.6% compared with the control group; the average prediction accuracy of the model of the present application scheme 2 is improved by 27.4% compared with the control group; the average prediction accuracy of the model of the present application scheme 2 is improved by 7.3% compared with the present application scheme 1. It can be seen that the model accuracy obtained by using the catalyst system relaxation energy model construction method disclosed in this application is higher than the model obtained by using the relevant technical solutions, among which the catalyst system relaxation energy model that has undergone multi-task training has higher accuracy.

[0127] FIG13 shows an exemplary structural block diagram of a device 1300 for constructing a catalyst system relaxation energy prediction model according to some embodiments of the present application.

[0128] As shown in FIG. 13 , the catalyst system relaxation energy prediction model construction device 1300 includes a first acquisition module 1310 , a first training module 1320 , a first construction module 1330 , a second acquisition module 1340 and a second training module 1350 .

[0129] The first acquisition module 1310 is used to acquire a first training data set, where the first training data set includes multiple first training data, each first training data includes a first training sample and a corresponding first sample label, wherein the first training sample includes first catalyst system structure information, and the first sample label includes system energy information corresponding to the first catalyst system structure information.

[0130] The first training module 1320 is configured to train the catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model.

[0131] The first constructing module 1330 is used to train the catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model.

[0132] The second acquisition module 1340 is used to acquire a second training data set, where the second training data set includes multiple second training data, each second training data includes a second training sample and a corresponding second sample label, wherein the second training sample includes second catalyst system structure information, and the second sample label includes relaxation energy information corresponding to the second catalyst system structure information.

[0133] The second training module 1350 is used to train the catalyst system relaxation energy initial prediction model using the second training data set to obtain the catalyst system relaxation energy prediction model.

[0134] Therefore, the catalyst system relaxation energy prediction model construction device 1300 can construct a catalyst system relaxation energy initial prediction model based on the pre-trained catalyst system energy prediction model, which makes it possible to quickly determine the catalyst system relaxation energy initial prediction model. Then, the second training module 1350 adjusts the catalyst system relaxation energy initial prediction model through the second training data set to obtain the catalyst system relaxation energy prediction model, which makes it possible to quickly obtain a catalyst system relaxation energy prediction model with high accuracy and fast response. It can be seen that using the catalyst system relaxation energy prediction model construction device 1300, a high-quality catalyst system relaxation energy prediction model can be quickly and easily constructed.

[0135] In some embodiments, the first sample tag further includes: atomic force information corresponding to the first catalyst system structure information, and the second sample tag further includes: atomic displacement information corresponding to the second catalyst system structure information.

[0136] In some embodiments, the first training module 1320 is also used to input the first training sample of each first training data in the first training data set into the catalyst system energy prediction model to obtain a first output result corresponding to the first training data; for each first training data in the first training data set, based on the first output result corresponding to the first training data and the first sample label of the first training data, calculate the first loss corresponding to the first training data; according to the first loss corresponding to each first training data in the first training data set, determine the first target loss of the catalyst system energy prediction model; based on the first target loss, iteratively update the parameters of the catalyst system energy prediction model until the first target loss meets the first preset condition to obtain a pre-trained catalyst system energy prediction model.

[0137] In some embodiments, when the first sample label also includes: atomic force information corresponding to the first catalyst system structure information, the first output result includes the catalyst system energy prediction result and the atomic force information prediction result; the first training module 1320 is also used to calculate the first sub-loss for each first training data in the first training data set based on the catalyst system energy prediction result corresponding to the first training data and the system energy information of the first training data; for each first training data in the first training data set, based on the atomic force information prediction result corresponding to the first training data and the atomic force information of the first training data, calculate the second sub-loss corresponding to the first training data; for each first training data in the first training data set, calculate the first sub-loss and the second sub-loss corresponding to the first training data.

[0138] In some embodiments, when the first sample label also includes: atomic force information corresponding to the first catalyst system structure information, the first output result includes the catalyst system energy prediction result and the atomic force information prediction result; the first training module 1320 is also used to calculate the second loss corresponding to each first training data in the first training data set based on the atomic force information prediction result corresponding to the first training data and the atomic force information of the first training data; determine the first target loss of the catalyst system energy prediction model based on the first loss and the second loss corresponding to each first training data in the first training data set.

[0139] In some embodiments, the first output result includes a catalyst system energy prediction result and an atomic force information prediction result; the catalyst system energy prediction model includes a topology map determination sub-model and a graph neural network sub-model, and the first training module 1320 is also used to input the first catalyst system structure information corresponding to each first training data in the first training data set into the topology map determination sub-model to obtain a topology map corresponding to the first catalyst system structure information, and the topology map is used to characterize the topological relationship between each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; for each first training data in the first training data set, the topology map corresponding to the first catalyst structure information corresponding to the first training data is input into the graph neural network sub-model to obtain a catalyst system energy prediction result and an atomic force information prediction result.

[0140] In some embodiments, the topological graph includes multiple nodes and edges connecting the multiple nodes in pairs, each node represents an atom in the corresponding catalyst system, and each edge represents the association between the two atoms corresponding to the edge; the first training module 1320 is also used to determine, for each first training data set, the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data according to the first catalyst system structure information corresponding to the first training data; for each first training data set, the position of each node in the topological graph corresponding to the first catalyst system structure information and the distance between each pair of atoms are determined according to the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; for each two-node combination in the topological graph, in response to the distance between the atoms corresponding to the two nodes in the combination being less than or equal to a predetermined distance threshold, the value of the edge connecting the two nodes is 1; for each two-node combination in the topological graph, in response to the distance between the atoms corresponding to the two nodes in the combination being greater than the predetermined distance threshold, the value of the edge connecting the two nodes is 0.

[0141] In some embodiments, the second training module 1350 is also used to input the second catalyst system structure information of each second training data in the second training data set into the catalyst system relaxation energy initial prediction model to obtain a second output result corresponding to the second training data; for each second training data in the second training data set, based on the second output result corresponding to the second training data and the second sample label, calculate the third loss corresponding to the second training data; determine the second target loss of the catalyst system relaxation energy initial prediction model based on the third loss corresponding to each second training data in the second training data set; based on the second target loss, iteratively update the parameters of the catalyst system relaxation energy initial prediction model until the second target loss meets the second preset condition to obtain the catalyst system energy prediction model.

[0142] In some embodiments, the second output result includes a catalyst system relaxation energy prediction result; the second training module 1350 is also used to calculate the third loss corresponding to each second training data in the second training data set based on the catalyst system relaxation energy prediction result corresponding to the second training data and the relaxation energy information of the second training data.

[0143] In some embodiments, the second sample label also includes: atomic displacement information corresponding to the second catalyst system structure information; the second output result also includes the atomic displacement information prediction result; the second training module 1350 is also used to calculate the fourth loss corresponding to each second training data in the second training data set based on the atomic displacement information prediction result corresponding to the second training data and the atomic displacement information of the second training data; determine the second target loss of the initial prediction model of the catalyst system relaxation energy based on the third loss and the fourth loss corresponding to each second training data in the second training data set.

[0144] In some embodiments, the first acquisition module 1310 is also used to obtain the three-dimensional structural information of each sample catalyst system in a plurality of sample catalyst systems, where the three-dimensional structural information includes the three-dimensional coordinates of each atom in the corresponding sample catalyst system; based on the three-dimensional structural information of each sample catalyst system in the plurality of sample catalyst systems, the system energy information and atomic force information of each sample catalyst system are determined by quantum mechanics methods; and the first training data is constructed according to the three-dimensional structural information of each sample catalyst system and its corresponding system energy information and atomic force information.

[0145] In some embodiments, the first training module 1320 is further configured to train the catalyst system energy prediction model using the first training data set based on a preset first hyperparameter set to obtain a pre-trained catalyst system energy prediction model, wherein the preset first hyperparameter set includes a preset first learning rate.

[0146] In some embodiments, the first training module 1320 is also used to train the catalyst system relaxation energy initial prediction model using a second training data set based on a preset second hyperparameter set to obtain a catalyst system relaxation energy prediction model, wherein the second hyperparameter set includes a preset second learning rate, and the second learning rate is less than the first learning rate.

[0147] In some embodiments, the second learning rate is equal to or less than one-tenth of the first learning rate.

[0148] As shown in FIG14 , the present application further provides a catalyst system relaxation energy prediction device 1400 , which includes a system structure information acquisition module 1410 and a model input module 1420 .

[0149] The system structure information acquisition module 1410 is used to acquire the system structure information of the catalyst system to be predicted.

[0150] The model input module 1420 is used to input the system structure information of the catalyst system to be predicted into the catalyst system relaxation energy prediction model obtained by the catalyst system relaxation energy prediction model construction method of any embodiment to obtain the relaxation energy information of the catalyst system to be predicted.

[0151] FIG15 illustrates an example system 1500 including an example computing device 1510 representing one or more systems and / or devices that can implement the various methods described herein. The computing device 1510 can be, for example, a server of a service provider, a device associated with a server, a system on a chip, and / or any other suitable computing device or computing system. The catalyst system relaxation energy prediction model building apparatus 1300 described above with reference to FIG9 can take the form of a computing device 1510. Alternatively, the catalyst system relaxation energy prediction model building apparatus 1300 can be implemented as a computer program in the form of an application 1516.

[0152] The example computing device 1510 as shown includes a processing system 1511, one or more computer-readable media 1512, and one or more I / O interfaces 1513 that are communicatively coupled to each other. Although not shown, the computing device 1510 may also include a system bus or other data and command transmission system that couples various components to each other. The system bus may include any one or a combination of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any one of a variety of bus architectures. Various other examples are also contemplated, such as control and data lines.

[0153] Processing system 1511 represents functionality that uses hardware to perform one or more operations. Thus, processing system 1511 is illustrated as including hardware elements 1514 that can be configured as processors, functional blocks, and the like. This can include hardware implemented as application-specific integrated circuits or other logic devices formed using one or more semiconductors. Hardware elements 1514 are not limited by the materials from which they are formed or the processing mechanisms employed therein. For example, a processor can be comprised of (a plurality of) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In such a context, processor-executable instructions can be electronically executable instructions.

[0154] Computer-readable media 1512 is illustrated as including memory / storage 1516. Memory / storage 1516 represents memory / storage capacity associated with one or more computer-readable media. Memory / storage 1516 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disks, magnetic disks, etc.). Memory / storage 1516 may include fixed media (e.g., RAM, ROM, fixed hard drive, etc.) and removable media (e.g., flash memory, removable hard drive, optical disks, etc.). Computer-readable media 1512 may be configured in various other ways as further described below.

[0155] One or more I / O interfaces 1513 represent functionality that allows a user to input commands and information to the computing device 1510 using various input devices, and optionally also allows information to be presented to the user and / or other components or devices using various output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone (e.g., for voice input), a scanner, touch functionality (e.g., a capacitive or other sensor configured to detect physical touch), a camera (e.g., that can detect motion that does not involve touch as gestures using visible or invisible wavelengths (such as infrared frequencies), etc.). Examples of output devices include a display device, a speaker, a printer, a network card, a tactile response device, etc. Thus, the computing device 1510 can be configured in various ways, as further described below, to support user interaction.

[0156] The computing device 1510 also includes an application 1516. The application 1516 may be, for example, a software instance of the catalyst system relaxation energy prediction model building apparatus 1300, and in combination with other elements in the computing device 1510 implements the techniques described herein.

[0157] The present application provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computing device to perform the catalyst system relaxation energy prediction model construction method provided in the various optional implementations described above.

[0158] Various techniques may be described herein in the general context of software, hardware, or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, and the like that perform specific tasks or implement specific abstract data types. As used herein, the terms "module," "function," and "component" generally refer to software, firmware, hardware, or a combination thereof. The techniques described herein are platform-independent, meaning that these techniques can be implemented on a variety of computing platforms with a variety of processors.

[0159] An implementation of the described modules and techniques may be stored on or transmitted across some form of computer-readable media. Computer-readable media may include various media accessible by computing device 1510. In some embodiments and without limitation, computer-readable media may include "computer-readable storage media" and "computer-readable signal media."

[0160] As opposed to a mere signal transmission, carrier wave, or signal itself, "computer-readable storage medium" refers to a medium and / or device, and / or tangible storage device, capable of persistently storing information. Thus, computer-readable storage media refers to non-signal-bearing media. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented in a method or technology suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage devices, hard disks, cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of manufacture suitable for storing desired information and accessible by a computer.

[0161] "Computer-readable signal media" refers to signal-bearing media that is configured to send instructions to the hardware of the computing device 1510, such as via a network. Signal media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism. Signal media also includes any information delivery media. The term "modulated data signal" refers to a signal that has one or more of its characteristics set or changed so as to encode information in the signal. In some embodiments, and not by way of limitation, communication media includes wired media such as a wired network or direct connection, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0162] As before, hardware elements 1514 and computer-readable medium 1512 represent instructions, modules, programmable device logic and / or fixed device logic implemented in hardware form, which in some embodiments can be used to implement at least some aspects of the technology described herein. Hardware elements can include other implementations in integrated circuits or systems on a chip, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs) and silicon or other hardware devices. In this context, hardware elements can be used as processing equipment for executing program tasks defined by the instructions, modules and / or logic embodied by the hardware elements, as well as hardware devices for storing instructions for execution, such as the computer-readable storage media described previously.

[0163] The aforementioned combination may also be used to implement the various techniques and modules herein. Therefore, software, hardware or program modules and other program modules may be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 1514. Computing device 1510 may be configured to implement specific instructions and / or functions corresponding to software and / or hardware modules. Therefore, for example, by using a computer-readable storage medium and / or hardware elements 1514 of a processing system, a module may be implemented as a module that can be executed by computing device 1510 as software, at least in part, in hardware. Instructions and / or functions may be executable / operable by one or more articles of manufacture (e.g., one or more computing devices 1510 and / or processing systems 1511) to implement the techniques, modules, and examples described herein.

[0164] In various embodiments, computing device 1510 can be implemented in a variety of different configurations. For example, computing device 1510 can be implemented as a computer-type device including a personal computer, a desktop computer, a multi-screen computer, a laptop computer, a netbook, etc. Computing device 1510 can also be implemented as a mobile device-type device including mobile devices such as mobile phones, portable music players, portable gaming devices, tablet computers, multi-screen computers, etc. Computing device 1510 can also be implemented as a television-type device, which includes devices having or connected to generally larger screens in casual viewing environments. These devices include televisions, set-top boxes, game consoles, etc.

[0165] The techniques described herein can be supported by these various configurations of computing device 1510 and are not limited to the specific examples of the techniques described herein. Functionality can also be implemented in whole or in part on the "cloud" 1520 using a distributed system, such as through platform 1522 as described below.

[0166] Cloud 1520 includes and / or represents a platform 1522 for resources 1524. Platform 1522 abstracts the underlying functionality of the hardware (e.g., servers) and software resources of cloud 1520. Resources 1524 may include applications and / or data that can be used when executing computer processing on servers remote from computing device 1510. Resources 1524 may also include services provided over the Internet and / or over a subscriber network such as a cellular or Wi-Fi network.

[0167] The platform 1522 can abstract resources and functionality to connect the computing device 1510 with other computing devices. The platform 1522 can also be used to abstract hierarchies of resources to provide a corresponding level of hierarchy in the demand encountered for resources 1524 implemented via the platform 1522. Thus, in an interconnected device embodiment, the implementation of the functionality described herein can be distributed throughout the system 1500. For example, functionality can be implemented partially on the computing device 1510 and through the platform 1522 that abstracts the functionality of the cloud 1520.

[0168] Should be understood that, for the sake of clarity, the embodiments of the present application are described with reference to different functional units. However, it will be apparent that, without departing from the present application, the functionality of each functional unit can be implemented in a single unit, implemented in multiple units or implemented as a part for other functional units. For example, the functionality that is described as being performed by a single unit can be performed by multiple different units. Therefore, reference to a specific functional unit is only considered as a reference to the appropriate unit for providing the described functionality, rather than indicating strict logical or physical structure or organization. Therefore, the application can be implemented in a single unit, or can be physically and functionally distributed between different units and circuits.

[0169] Although the present application has been described in conjunction with some embodiments, it is not intended to be limited to the specific forms set forth herein. On the contrary, the scope of the present application is limited only by the appended claims. Additionally, although individual features may be included in different claims, these may possibly be advantageously combined, and inclusion in different claims does not imply that a combination of features is not feasible and / or advantageous. The order of the features in the claims does not imply any specific order in which the features must work. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the term "a" or "an" does not exclude a plurality. The reference numerals in the claims are provided merely as clear examples and should not be construed as limiting the scope of the claims in any way.

[0170] It is understood that in the specific implementation of this application, data related to software test cases, etc. is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of relevant data must comply with relevant laws, regulations, and standards of relevant countries and regions.

[0171] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for constructing a catalyst system relaxation energy prediction model, executed by a computing device, the method comprising: Acquire a first training data set, wherein the first training data set includes a plurality of first training data, each of the first training data includes a first training sample and a corresponding first sample label, wherein the first training sample includes first catalyst system structure information, and the first sample label includes system energy information corresponding to the first catalyst system structure information; Using the first training data set to train a catalyst system energy prediction model to obtain a pre-trained catalyst system energy prediction model; Constructing an initial prediction model for catalyst system relaxation energy based on the pre-trained catalyst system energy prediction model; Acquire a second training data set, wherein the second training data set includes a plurality of second training data, each of the second training data includes a second training sample and a corresponding second sample label, wherein the second training sample includes second catalyst system structure information, and the second sample label includes relaxation energy information corresponding to the second catalyst system structure information; and The catalyst system relaxation energy initial prediction model is trained using the second training data set to obtain a catalyst system relaxation energy prediction model.

2. The method according to claim 1, wherein the first sample label further comprises: The atomic force information corresponding to the first catalyst system structure information, the second sample label also includes: atomic displacement information corresponding to the second catalyst system structure information.

3. The method according to claim 1 or 2, wherein the using the first training data set to train the catalyst system energy prediction model to obtain a pre-trained catalyst system energy prediction model comprises: For each first training data in the first training data set, inputting a first training sample of the first training data into the catalyst system energy prediction model to obtain a first output result corresponding to the first training data; For each first training data in the first training data set, based on a first output result corresponding to the first training data and a first sample label of the first training data, calculating a first loss corresponding to the first training data; Determining a first target loss of the catalyst system energy prediction model according to a first loss corresponding to each first training data in the first training data set; Based on the first target loss, the parameters of the catalyst system energy prediction model are iteratively updated until the first target loss meets a first preset condition, so as to obtain a pre-trained catalyst system energy prediction model.

4. The method according to claim 3, wherein the first sample label further comprises: Atomic force information corresponding to the first catalyst system structure information, the first output result comprising a catalyst system energy prediction result and an atomic force information prediction result; The calculating, for each first training data in the first training data set, a first loss corresponding to the first training data based on a first output result corresponding to the first training data and a first sample label of the first training data comprises: For each first training data in the first training data set, calculating a first sub-loss based on a catalyst system energy prediction result corresponding to the first training data and system energy information of the first training data; For each first training data in the first training data set, based on the atomic force information prediction result corresponding to the first training data and the atomic force information of the first training data, calculate the second sub-loss corresponding to the first training data; For each first training data in the first training data set, a first sub-loss and a second sub-loss corresponding to the first training data are determined.

5. The method according to claim 3 or 4, wherein the first sample label further comprises: Atomic force information corresponding to the first catalyst system structure information, the first output result comprising a catalyst system energy prediction result and an atomic force information prediction result; The determining, according to the first loss corresponding to each first training data in the first training data set, the first target loss of the catalyst system energy prediction model comprises: For each first training data in the first training data set, based on the atomic force information prediction result corresponding to the first training data and the atomic force information of the first training data, calculate the second loss corresponding to the first training data; A first target loss of the catalyst system energy prediction model is determined according to the first loss and the second loss corresponding to each first training data in the first training data set.

6. The method according to any one of claims 3 to 5, wherein the first output result includes a catalyst system energy prediction result and an atomic force information prediction result; the catalyst system energy prediction model includes a topology map determination submodel and a graph neural network submodel, and for each first training data in the first training data set, the first training sample of the first training data is input into the catalyst system energy prediction model to obtain a first output result corresponding to the first training data, including: For each first training data in the first training data set, inputting the first catalyst system structure information corresponding to the first training data into the topology map determination submodel to obtain a topology map corresponding to the first catalyst system structure information, wherein the topology map is used to characterize the topological association between each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; For each first training data in the first training data set, the topological map corresponding to the first catalyst structure information corresponding to the first training data is input into the graph neural network sub-model to obtain the catalyst system energy prediction results and the atomic force information prediction results.

7. The method according to claim 6, wherein the topological graph comprises a plurality of nodes and edges connecting the plurality of nodes in pairs, each node represents an atom in the corresponding catalyst system, and each edge represents a relationship between two atoms corresponding to the edge; The step of inputting the first catalyst system structure information corresponding to each first training data in the first training data set into the topology map determination submodel to obtain a topology map corresponding to the first catalyst system structure information includes: For each first training data in the first training data set, according to the first catalyst system structure information corresponding to the first training data, determine the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; For each first training data in the first training data set, according to the three-dimensional coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data, determine the position of each node of the topological graph corresponding to the first catalyst system structure information and the distance between each atom; For each dual-node combination in the topological graph, in response to the distance between atoms corresponding to the two nodes in the combination being less than or equal to a predetermined distance threshold, the value of the edge connecting the two nodes is 1; For each dual-node combination in the topological graph, in response to the distance between atoms corresponding to two nodes in the combination being greater than a predetermined distance threshold, the value of the edge connecting the two nodes is 0.

8. The method according to any one of claims 1 to 7, wherein the using the second training data set to train the catalyst system relaxation energy initial prediction model to obtain the catalyst system relaxation energy prediction model comprises: For each second training data in the second training data set, inputting the second catalyst system structure information of the second training data into the catalyst system relaxation energy initial prediction model to obtain a second output result corresponding to the second training data; For each second training data in the second training data set, based on the second output result and the second sample label corresponding to the second training data, calculate the third loss corresponding to the second training data; Determining a second target loss of the catalyst system relaxation energy initial prediction model according to a third loss corresponding to each second training data in the second training data set; Based on the second target loss, the parameters of the catalyst system relaxation energy initial prediction model are iteratively updated until the second target loss meets a second preset condition, so as to obtain a catalyst system energy prediction model.

9. The method according to claim 8, wherein the second output result comprises a catalyst system relaxation energy prediction result; The calculating, for each second training data in the second training data set, the third loss corresponding to the second training data based on the second output result and the second sample label corresponding to the second training data includes: For each second training data in the second training data set, based on the catalyst corresponding to the second training data The system relaxation energy prediction result and the relaxation energy information of the second training data are used to calculate the third loss corresponding to the second training data.

10. The method according to claim 9, wherein the second sample label further comprises: Atomic displacement information corresponding to the second catalyst system structure information; The second output result also includes a prediction result of atomic displacement information; The determining, according to the third loss corresponding to each second training data in the second training data set, the second target loss of the initial prediction model of the catalyst system relaxation energy comprises: For each second training data in the second training data set, based on the atomic displacement information prediction result corresponding to the second training data and the atomic displacement information of the second training data, calculating a fourth loss corresponding to the second training data; The second target loss of the catalyst system relaxation energy initial prediction model is determined according to the third loss and the fourth loss corresponding to each second training data in the second training data set.

11. The method according to any one of claims 1 to 10, wherein obtaining the first training data set comprises: Acquire three-dimensional structural information of each sample catalyst system in a plurality of sample catalyst systems, the three-dimensional structural information including three-dimensional coordinates of each atom in the corresponding sample catalyst system; Based on the three-dimensional structure information of each of the plurality of sample catalyst systems, determining the system energy information and atomic force information of each sample catalyst system by quantum mechanics method; The first training data is constructed according to the three-dimensional structure information of each sample catalyst system and its corresponding system energy information and atomic force information.

12. The method according to any one of claims 1 to 11, wherein the step of training a catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model comprises: Based on a preset first hyperparameter set, the catalyst system energy prediction model is trained using the first training data set to obtain a pre-trained catalyst system energy prediction model, wherein the preset first hyperparameter set includes a preset first learning rate, Furthermore, the step of training the catalyst system relaxation energy initial prediction model using the second training data set to obtain the catalyst system relaxation energy prediction model comprises: Based on a preset second hyperparameter set, the catalyst system relaxation energy initial prediction model is trained using the second training data set to obtain a catalyst system relaxation energy prediction model, wherein the second hyperparameter set includes a preset second learning rate, and the second learning rate is less than the first learning rate.

13. The method according to claim 12, wherein the second learning rate is equal to or less than one tenth of the first learning rate.

14. A method for predicting relaxation energy of a catalyst system, the method comprising: Acquiring system structure information of the catalyst system to be predicted; The system structure information of the catalyst system to be predicted is input into the catalyst system relaxation energy prediction model according to any one of claims 1 to 13 to obtain the relaxation energy information of the catalyst system to be predicted.

15. A catalyst system relaxation energy prediction model construction device, comprising: A first acquisition module is used to acquire a first training data set, wherein the first training data set includes a plurality of first training data, each of the first training data includes a first training sample and a corresponding first sample label, wherein the first training sample includes first catalyst system structure information, and the first sample label includes system energy information corresponding to the first catalyst system structure information; A first training module, used for training a catalyst system energy prediction model using the first training data set to obtain a pre-trained catalyst system energy prediction model; A first construction module is used to construct an initial prediction model of catalyst system relaxation energy based on the pre-trained catalyst system energy prediction model; A second acquisition module is used to acquire a second training data set, wherein the second training data set includes a plurality of second training data, each of the second training data includes a second training sample and a corresponding second sample label, wherein the second training sample includes second catalyst system structure information, and the second sample label includes relaxation energy information corresponding to the second catalyst system structure information; A second training module is used to train the catalyst system relaxation energy initial prediction using the second training data set. Model is used to obtain a catalyst system relaxation energy prediction model.

16. A computing device comprising: a memory configured to store computer-executable instructions; as well as A processor configured to perform the method according to any one of claims 1-14 when the computer executable instructions are executed by the processor.

17. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions implement the method according to any one of claims 1 to 14 when executed.

18. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 14.

Images