Method and apparatus for constructing a catalytic system relaxation energy prediction model, computing equipment and computer program

Abstract

A method for constructing a catalyst system relaxation energy prediction model, which is performed by a computer, is provided. This method includes the steps of: obtaining a first training dataset, the first training dataset containing multiple first training data, each first training data containing a first training sample and a corresponding first sample label, the first training sample containing first catalyst system structure information, and the first sample label containing system energy information corresponding to the first catalyst system structure information (S310); obtaining a pre-trained catalyst system energy prediction model by training a catalyst system energy prediction model using the first training dataset (S320); constructing an initial catalyst system relaxation energy prediction model based on the pre-trained catalyst system energy prediction model (S330); obtaining a second training dataset, the second training dataset containing multiple second training data, each second training data containing a second training sample and a corresponding second sample label, the second training sample containing second catalyst system structure information, and the second sample label containing relaxation energy information corresponding to the second catalyst system structure information (S340); and obtaining a catalyst system relaxation energy prediction model by training an initial catalyst system relaxation energy prediction model using the second training dataset (S350).

Description

Technical Field

[0003]

[0001] This application claims priority based on a Chinese patent application filed with the Chinese Patent Office on September 14, 2023, with an application number of 2023111904865 and an invention title of "Method and Apparatus for Constructing a Catalyst System Relaxation Energy Prediction Model", and a Chinese patent application filed with the Chinese Patent Office on September 14, 2023, with an application number of 2023111901833 and an invention title of "Method and Apparatus for Constructing a Catalyst System Relaxation Energy Prediction Model", and incorporates all of their contents herein by reference.

[0002] This application relates to the technical field of computers, and particularly to a method and apparatus for constructing a catalyst system relaxation energy prediction model, a computing device, and a computer program.

Background Art

[0003] As an important auxiliary means for chemical reactions, catalysts are widely used in fields such as the chemical industry and manufacturing. Selecting an appropriate catalyst for a chemical reaction is an important means to control the reaction rate and ensure reaction safety. In conventional methods, generally, a large number of experiments are used to select an appropriate catalyst from a huge amount of alternative substances. Such a method not only takes time and effort but also causes extremely large waste. Therefore, it is desirable for experimenters to pre-predict some important indicators of the catalyst to reduce the options during catalyst selection and thus achieve the rapid discovery of an appropriate catalyst. The relaxation energy can indicate the energy that an adsorbate-catalyst system can release from the initial state to the stable state, and it is an important indicator for evaluating the performance of the catalyst. Conventional quantum mechanics-based relaxation energy prediction methods have a high computational complexity and a huge amount of calculations, which can seriously affect the efficiency of catalyst research and development. In recent years, with the development of computer technology, using deep learning methods to construct a relaxation energy prediction model to predict the relaxation energy of an adsorbate-catalyst system has been attracting increasing attention. [Overview of the project] [Problems that the invention aims to solve]

[0004] This application aims to provide a method and apparatus for constructing a catalyst system relaxation energy prediction model, a computing device, and a computer program. [Means for solving the problem]

[0005] According to one aspect of this application, a method for constructing a catalytic system relaxation energy prediction model is provided, which is performed by a computer, and such method is A first training dataset is obtained, the first training dataset includes multiple first training data, each first training data includes a first training sample and a corresponding first sample label, 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 pre-trained catalytic system energy prediction model is obtained by training the catalytic system energy prediction model using the aforementioned first training dataset; Based on the aforementioned pre-training catalyst system energy prediction model, an initial prediction model for catalyst system relaxation energy is constructed; A second training dataset is obtained, the second training dataset includes a plurality of second training data, each second training data includes a second training sample and a corresponding second sample label, of which 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 process includes the step of obtaining a catalytic system relaxation energy prediction model by training the initial prediction model of the catalytic system relaxation energy using the second training dataset.

[0006] According to another aspect of this application, a device for constructing a catalytic system relaxation energy prediction model is provided, and such device is A first acquisition module used to acquire a first training dataset, wherein the first training dataset includes a plurality of first training data sets, each first training data set includes a first training sample and a corresponding first sample label, 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 to obtain a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the aforementioned first training dataset; A first construction module used to construct an initial prediction model of the catalytic system relaxation energy based on the aforementioned pre-trained catalytic system energy prediction model; A second acquisition module used to acquire a second training dataset, wherein the second training dataset includes a plurality of second training data sets, each second training data set includes a second training sample and a corresponding second sample label, of which 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 This includes a second training module used to obtain a catalytic system relaxation energy prediction model by training the catalytic system relaxation energy prediction model using the second training dataset.

[0007] According to another aspect of this application, a method for predicting the relaxation energy of a catalyst system is provided, and such method is Obtain the system structure of the predictive catalyst system; and The method includes the step of obtaining the relaxation energy of the catalyst system by inputting the system structure of the catalyst system awaiting prediction into the catalyst system relaxation energy prediction model in any one of the embodiments described above.

[0008] According to another aspect of the present application, a computing device is provided which includes a memory for storing computer-executable instructions; and a processor connected to the memory, the processor configured to perform steps of a method for constructing a catalytic system relaxation energy prediction model in some embodiments of the present application when the computer-executable instructions are executed by the processor.

[0009] According to another aspect of this application, a computer-readable storage medium is provided which stores computer-executable instructions, and when these computer-executable instructions are executed, steps of a method for constructing a catalytic system relaxation energy prediction model in some embodiments of this application are realized.

[0010] According to another aspect of this application, a computer program product is provided which includes a computer program that, when executed by a processor, realizes the steps of a method for constructing a catalytic system relaxation energy prediction model in some embodiments of this application.

[0011] The embodiments described below will clarify these and other advantages of this application, and these and other advantages of this application will be explained in conjunction with the embodiments described below. [Brief explanation of the drawing]

[0012] To more clearly describe the embodiments of this application or the technical concepts in the prior art, the drawings that are necessary for describing the embodiments or the prior art are briefly introduced below. Clearly, the drawings described below are merely embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any creative effort. [Figure 1] This is a flowchart of the catalyst research and development process in the embodiment of this application. [Figure 2] This figure shows an exemplary application scenario of the catalyst system relaxation energy prediction model construction method in the embodiment of this application. [Figure 3]This is a flowchart of the method for constructing a catalyst system relaxation energy prediction model in an embodiment of this application. [Figure 4] This figure shows how to obtain a pre-trained catalyst system energy prediction model by training a catalyst system energy prediction model using a first training dataset in an embodiment of this application. [Figure 5] This figure shows how to train a catalyst system energy prediction model in an embodiment of this application. [Figure 6] This is a flowchart for obtaining a catalyst system relaxation energy prediction model by training an initial prediction model of the catalyst system relaxation energy using the second training dataset. [Figure 7] This figure shows how to train a catalyst system relaxation energy prediction model in an embodiment of this application. [Figure 8] This figure shows how to train a catalytic system relaxation energy prediction model for multitasking in an embodiment of this application. [Figure 9] This figure shows how to train a catalyst system relaxation energy prediction model in an embodiment of this application. [Figure 10] This figure shows a method for constructing a catalyst system relaxation energy prediction model, which is pre-trained but not multi-task trained, as described in the embodiment of this application. [Figure 11] This figure shows a method for constructing a catalytic system relaxation energy prediction model for multitasking in an embodiment of this application. [Figure 12] This invention provides an example of a catalyst system relaxation energy prediction model and a comparison diagram of the effects of a control group. [Figure 13] This is a block diagram of an exemplary structure of a catalyst system relaxation energy prediction model construction apparatus in one embodiment of the present application. [Figure 14] This figure shows the structure of a catalyst system relaxation energy prediction device in one embodiment of the present application. [Figure 15]A diagram showing an exemplary system including an exemplary computer device representative of one or more systems and / or devices capable of implementing the various methods described in this application.

Best Mode for Carrying Out the Invention

[0013] Hereinafter, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the drawings in the embodiments of this application. As is clear, the described embodiments are only some of the embodiments of this application, not all of them. All other embodiments obtained by those skilled in the art without creative efforts based on the embodiments in this application shall fall within the protection scope of this application. In the drawings, the same reference numerals represent the same or similar parts, so the repeated descriptions thereof are omitted here.

[0014] Also, the features, structures or characteristics described can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to give a full understanding of the embodiments of this application. However, as can be understood by those skilled in the art, one or more of these specific details may not be necessary to practice the technical solutions of this application, or other methods, assemblies, devices, steps, etc. may be adopted. In other cases, well-known methods, devices, implementations or operations are not disclosed or described in detail to avoid obscuring each aspect of this application.

[0015] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be realized in the form of software, or may be realized by one or more hardware modules or integrated circuits, or may be realized by different networks and / or processors and / or microcontrollers.

[0016] It should be noted that while various assemblies may be described here using terms such as first, second, and third, these assemblies should not be limited to these terms. These terms are still used to distinguish one assembly from another. Therefore, for example, if the first assembly described below is referred to as the second assembly, it does not deviate from teaching the concepts of this application. Furthermore, the terms “and / or” and similar terms used herein include any one, more, and all combinations of the items listed in association. Before describing the embodiments of this application in detail, we will first briefly introduce some of the terms used in the embodiments of this application so that those skilled in the art can understand them.

[0017] Catalyst: A catalyst is generally a substance that improves the reaction rate without changing the standard Gibbs free energy of the entire reaction. Generally, a catalyst is a substance that participates in an intermediate step of a chemical reaction and can selectively change the reaction rate, but whose quantity and chemical properties do not substantially change before and after the reaction. Alternatively, it can be described as a substance that improves the reaction rate without changing the chemical equilibrium in a chemical reaction, and whose mass and chemical properties do not change before and after the reaction. Statistics show that catalysts are used in more than 90% of industrial processes in fields such as chemistry, petrochemicals, biochemistry, and environmental protection. Catalysts are diverse and can be classified into liquids and solids based on their state, and also into homogeneous and multiphase catalysts based on the phase of the reaction system. Homogeneous catalysts include acids, bases, soluble transition metal compounds, and peroxides. Catalysts play a crucial role in the modern chemical industry; for example, iron catalysts are used in the production of synthetic ammonia, vanadium catalysts in the production of sulfuric acid, and different catalysts are used in the production of three major synthetic materials, such as ethylene polymerization and butadiene rubber production. Finding catalysts suitable for chemical reactions is a constant challenge in the chemical industry, and it is also an important means of controlling reaction rates and improving reaction safety.

[0018] A catalytic system, also called an adsorbent-catalyst system, refers to a system consisting of multiple atoms or molecules acting as a catalyst, and may also include atoms or molecules as reactants adsorbed onto the catalyst. Because a catalytic system contains multiple atoms or molecules, and the reactants acting as adsorbents continue to react, the forces between atoms or molecules constantly change. Therefore, the system's energy also changes until the reaction is complete and the system moves towards a steady state. Thus, each microstructure of a catalytic system corresponds to one system energy, and at this time, there is a corresponding force relationship between the atoms or molecules within it. Typically, the difference between the system energy corresponding to the initial structure of a catalytic system and the system energy corresponding to the final structure after the reaction is complete is called the system's relaxation energy, and is used to evaluate the performance of the catalyst. The larger the catalytic system's relaxation energy, the better the catalyst's performance.

[0019] Graph Neural Networks (GNNs) are a general term for algorithms that use neural networks to learn from graph-structured data, extract and mine features and patterns in graph-structured data, and satisfy the needs of graph learning tasks such as clustering, classification, prediction, partitioning, and generation. By creating specific policies for nodes and edges in a graph, GNNs transform graph-structured data into normative and standardized representations, and then train various neural networks with this data, achieving excellent results in tasks such as node classification, edge information propagation, and graph clustering.

[0020] Topological Maps: Topological maps simplify and standardize the relationships between entities and display quantized information based on them. Topological maps are an effective representation of relationships between entities by conveying quantized information in a graph.

[0021] Figure 1 shows flowcharts of catalyst research and development in several embodiments of this application. As shown in Figure 1, catalysts are typically screened efficiently using computing equipment 110 and experimental equipment 120 during catalyst research and development. First, relaxation energies are predicted for a large number of catalysts using computing equipment, and catalysts with relatively good prediction results are selected. Next, experimental verification is performed on the selected catalysts using experimental equipment 120, and catalysts whose experimental performance satisfies the expected value are selected. While this catalyst screening method using computing equipment can greatly improve the efficiency of catalyst research and development, the accuracy and speed of the computing equipment 110 in predicting relaxation energies are high. Therefore, it is necessary to construct a catalyst system relaxation energy prediction model. Catalyst systems have complex atomic structures and rapid structural changes during chemical reactions, making the construction of a relaxation energy prediction model difficult and resulting in low accuracy. In related technical proposals, models can be constructed using quantum mechanical methods, but these methods have many drawbacks. Quantum mechanical models typically use density functional theory (DENSI) methods to perform molecular simulations, and then predict the relaxation energy of catalytic systems based on these simulations. However, both the construction and operation of such models require calculations for each molecule or atom, resulting in extremely high computational costs and slow processing speeds that fail to meet the needs of practical applications.

[0022] Therefore, this application provides a method for constructing a catalyst system relaxation energy prediction model. The model is executed on a computing device 110 to achieve accurate predictions of the catalyst system relaxation energy.

[0023] Figure 2 shows an exemplary application scenario 200 of a method for constructing a catalytic system relaxation energy prediction model in several embodiments of the present application. The application scenario 200 may include a server 210, terminal equipment 220, and server 230. Servers 210, terminal equipment 220, and server 230 are connected communicatively via a network 240. The 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, and any other type of network known to those skilled in the art.

[0024] In some embodiments, the method for constructing a catalyst system relaxation energy prediction model may be mainly performed on server 210. Server 210 acquires a first training dataset, which includes multiple first training data sets, each containing a first training sample and a corresponding first sample label, where the first training sample contains first catalyst system structure information, and the first sample label contains system energy information corresponding to the first catalyst system structure information. In some embodiments, the first training dataset may be stored on server 210 or acquired from other servers or terminals via network 240. Subsequently, server 210 acquires a pre-trained catalyst system energy prediction model by training a catalyst system energy prediction model using the first training dataset, and then constructs an initial catalyst system relaxation energy prediction model based on the pre-trained catalyst system energy prediction model. Subsequently, server 210 acquires a second training dataset, which includes multiple second training data sets, each containing a second training sample and a corresponding second sample label, where the second training sample contains second catalyst system structure information, and the second sample label contains relaxation energy information corresponding to the second catalyst system structure information. In some embodiments, the second training dataset may be stored on server 210, or it may be retrieved from other servers or terminals via network 240. Finally, server 210 obtains a catalytic system relaxation energy prediction model by training the catalytic system relaxation energy initial prediction model using the second training dataset.

[0025] In some embodiments, the method for constructing a catalyst system relaxation energy prediction model may be mainly performed on terminal equipment 220 or server 230. Server 210, terminal equipment 220, and server 230 may each include media and / or devices capable of non-temporarily storing information, and / or tangible memory devices. Therefore, computer-readable storage media refers to non-signal carrier media. Computer-readable storage media include, for example, volatile and non-volatile, removable and non-removable media, and / or hardware such as memory devices implemented by methods or techniques suitable for storing information (e.g., computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). As those skilled in the art will understand, an example of server 210 may be an independent physical server, a group of multiple physical servers, or a distributed system, and may also be 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, big data, and artificial intelligence platforms. The terminal and the server may be connected directly or indirectly by wired or wireless means, but this application is not limited thereto. The server 210 presents the data allocation policy awaiting determination by the terminal device 220 to the developer and interacts with the developer to make the development policy visible.

[0026] The terminal device 220 may be any type of mobile computing device, such as 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, smart glasses, or other head-mounted devices), or other types of mobile devices. In some embodiments, the terminal device 220 and server 230 may be fixed computing devices, such as a desktop computer, game console, smart TV, etc. Furthermore, if the application scenario 200 includes multiple terminal devices 220, these multiple terminal devices 220 may be the same or different types of computing devices.

[0027] As shown in Figure 2, the terminal device 220 may include a display screen and a terminal application that interacts with the terminal user via the display screen. The terminal application may be a local application program, a web application program, or a lightweight application such as a LiteApp (e.g., a mobile app or a WeChat app). If the terminal application is a local application program that requires installation, the terminal application can be installed on the terminal device 220. If the terminal application is a web application program, it can be accessed via a browser. If the terminal application is a Lite program, the terminal application can be opened directly on the terminal device 220 without installation by searching for information related to the terminal application (e.g., the name of the terminal application) and scanning the graphic code of the terminal application (e.g., a barcode, QR code).

[0028] In some embodiments, the application scenario 200 described above may be a distributed system consisting of servers 230, which can constitute, for example, a blockchain system. Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and cryptographic algorithms. A blockchain is essentially a distributed database, a series of data blocks generated using cryptographic techniques. Each data block contains information about a batch of network transactions for the purpose of verifying the validity (anti-counterfeiting) of that information and generating the next block. A blockchain may include a platform layer, a platform product service layer, and an application service layer.

[0029] Figure 3 is an exemplary flowchart of the method 400 for constructing a catalytic system relaxation energy prediction model in some embodiments of the present application. As shown in Figure 3, the method 300 includes steps S310, S320, S330, S340 and S350.

[0030] In step S310, a first training dataset is acquired, which includes multiple first training data sets, each of which includes a first training sample and a corresponding first sample label, the first training sample containing first catalyst system structure information, and the first sample label containing system energy information corresponding to the first catalyst system structure information.

[0031] Of these, the first catalyst system structure information is the structure information of the catalyst system represented by the first training sample, and can represent the structure of the catalyst system. The system energy information corresponding to the first catalyst system structure information can represent the system energy of the catalyst system represented by the first catalyst system structure information.

[0032] In some embodiments, system energy information corresponding to the first catalyst system structure information can be acquired and stored by measurement or quantum mechanical calculation for retrieval during use. In other embodiments, the first sample label may further include interatomic force information corresponding to the first catalyst system structure information. Interatomic force information can be acquired and stored by measurement or quantum mechanical calculation for retrieval during use.

[0033] In step S320, a pre-trained catalyst system energy prediction model is obtained by training the catalyst system energy prediction model using the first training dataset.

[0034] In some embodiments, the catalyst system energy prediction model may include a graph neural network, and training the catalyst system energy prediction model using the first training dataset involves training the parameters of the graph neural network using the first training dataset to obtain a graph neural network capable of predicting the catalyst system energy based on the catalyst system structure.

[0035] In step S330, an initial prediction model for the catalytic system relaxation energy is constructed based on a pre-trained prediction model for the catalytic system energy.

[0036] In some embodiments, a pre-trained catalyst system energy prediction model can be used as the initial catalyst system relaxation energy prediction model. Because the pre-trained catalyst system energy prediction model is a multi-task model (the model simultaneously predicts the system energy and the forces acting on the catalyst system based on the catalyst system structure), the parameter tuning and determination process is more rigorous than training a single-task model (for example, it can be more comprehensive during loss calculation). This allows for more efficient parameter tuning and determination, faster convergence, and greater adaptability of the acquired parameters.

[0037] Since the output of the initial catalytic system relaxation energy prediction model includes outputs similar to those of the initial catalytic system relaxation energy prediction model (e.g., all including one-dimensional and matrix outputs), using the pre-trained catalytic system energy prediction model as the initial catalytic system relaxation energy prediction model allows us to determine one relatively good initial parameter value for subsequent model construction.

[0038] In step S340, a second training dataset is acquired, the second training dataset containing multiple second training data sets, each second training data set containing a second training sample and a corresponding second sample label, the second training sample containing second catalyst system structure information, and the second sample label containing relaxation energy information corresponding to the second catalyst system structure information.

[0039] In some embodiments, the second catalyst system structure information is the structure information of the catalyst system represented by the second training sample, and can represent the structure of the catalyst system. The relaxation energy information corresponding to the second catalyst system structure information can represent the relaxation energy of the catalyst system represented by the second catalyst system structure information.

[0040] In some embodiments, the relaxation energy corresponding to the second catalyst system structure information can be acquired and stored by measurement or quantum mechanical calculation for recall during use. The second sample label may further 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 acquired and stored by measurement or quantum mechanical calculation for recall during use.

[0041] In step S350, a catalytic system relaxation energy prediction model is obtained by training an initial prediction model of the catalytic system relaxation energy using the second training dataset.

[0042] In some embodiments, the initial prediction model for the catalytic system relaxation energy may include a graph neural network, and the initial prediction model for the catalytic system relaxation energy may be trained using a second training dataset, or further training may be performed using the second training dataset based on parameters pre-trained by the graph neural network, thereby obtaining a graph neural network capable of predicting the catalytic system relaxation energy based on the catalytic system structure.

[0043] Therefore, Method 400 constructs an initial catalyst system relaxation energy prediction model based on a pre-trained catalyst system energy prediction model, thereby enabling rapid determination of the initial catalyst system relaxation energy prediction model. Subsequently, the catalyst system relaxation energy prediction model is obtained by adjusting the initial catalyst system relaxation energy prediction model with a second training dataset, thereby enabling rapid acquisition of a highly accurate and fast-responding catalyst system relaxation energy prediction model. As can be seen, on the one hand, the catalyst system relaxation energy prediction model constructed by Method 400 employs a two-stage training process, significantly improving the accuracy of relaxation energy prediction and thus solving the problem of low computational accuracy in related technologies. On the other hand, the catalyst system relaxation energy prediction models constructed in some embodiments of this application can directly predict relaxation energy based on initial system structure information, eliminating the need for iterative calculations and resulting in fast computation speeds. This avoids the problems of high computational complexity, cumbersome computation processes, and slow computation speeds associated with conventional quantum mechanics methods and related technologies. Therefore, Method 400 enables the rapid and easy construction of high-quality catalyst system relaxation energy prediction models.

[0044] In some embodiments, the first sample label may further include interatomic force information corresponding to the structural information of the first catalyst system, and the second sample label may further include atomic displacement information corresponding to the structural information of the second catalyst system.

[0045] Of these, the interatomic force information represents information about interatomic forces in a catalyst system having the system structure represented by the first catalyst system structure information, and may include the magnitude and direction of the force. The atomic displacement information represents information about atomic displacement in the catalyst system. The atomic displacement of a certain atom in the catalyst system may 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 may include the magnitude and direction of the atomic displacement.

[0046] In some embodiments, the second training dataset contains various expected prediction data (e.g., relaxation energy and atomic displacement), allowing the model trained using the second dataset to handle multiple tasks simultaneously. Furthermore, since the initial prediction model for catalytic system relaxation energy is a multi-task model, rapid convergence and high model prediction accuracy can be achieved by adjusting the model's parameters using the second training data.

[0047] In this embodiment, the initial catalytic system relaxation energy prediction model for multitasks can be rapidly determined by constructing an initial catalytic system relaxation energy prediction model for multitasks based on a pre-trained catalytic system energy prediction model for multitasks. Subsequently, the catalytic system relaxation energy prediction model for multitasks can be obtained by adjusting the initial prediction model for multitasks with a second training dataset. Since both the first and second stage training trains the model to adapt to multitasks (e.g., the first stage includes two tasks: predicting system energy and interatomic force, and the second stage includes two tasks: predicting relaxation energy and atomic displacement), and the dimensions of the model's expected output correspond in the first and second stages (e.g., the data dimensions of system energy and relaxation energy are the same, and the data dimensions of interatomic force and atomic displacement are the same), the second stage training process is faster and more efficient, and the prediction accuracy of the final catalytic system relaxation energy prediction model is also higher. As can be seen, on the one hand, the catalytic system relaxation energy prediction model employs a two-stage training process, which significantly improves the accuracy of relaxation energy prediction, thus solving the problem of low computational accuracy in related technologies. On the other hand, the catalytic system relaxation energy prediction model can directly predict relaxation energy based on the initial system structure information, does not require iterative calculations, and is computationally fast, thus avoiding the problems of high computational complexity, cumbersome computation processes, and slow computation speed of conventional quantum mechanics methods and related technologies. Furthermore, since the catalytic system relaxation energy prediction model is trained on multiple tasks in both the first and second stages, its training efficiency is higher, and the resulting model prediction accuracy is better.

[0048] Figure 4 is a flowchart for obtaining a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the first training dataset. As shown in Figure 4, in some embodiments, step S320 includes steps S410, S420, S430 and S440.

[0049] In step S410, for each first training data in the first training dataset, the first training sample of the first training data is input into the catalyst system energy prediction model to obtain the first output result corresponding to the first training data.

[0050] In step S420, for each first training data in the first training dataset, a first loss corresponding to the first training data is calculated based on the first output result corresponding to the first training data and the first sample label of the first training data. In some embodiments, step S420 includes the following: for each first training data in the first training dataset, a first loss corresponding to the first training data is calculated based on the catalyst system energy prediction result corresponding to the first training data and the system energy information of the first training data.

[0051] In step S430, the 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 dataset. In some embodiments, the first target loss can be calculated using a loss function.

[0052] In some embodiments, when the first sample label further includes interatomic force information corresponding to first catalyst system structure information, and the first output result includes catalyst system energy prediction results and interatomic force information prediction results, step S430 includes the following: for each first training data in the first training dataset, calculate the second loss corresponding to the first training data based on the interatomic force information prediction result and the interatomic force information of the first training data; and determine 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 dataset. Of these, when calculating the first target loss, the sum of all first losses and all second losses obtained by the calculation can be obtained, and the result of the obtained sum can be taken as the first target loss.

[0053] In step S440, a pre-trained catalyst system energy prediction model is obtained by iteratively updating the parameters of the catalyst system energy prediction model based on the first target loss until the first target loss satisfies the first predetermined condition. Thus, the pre-trained catalyst system energy prediction model can accurately predict the catalyst system energy based on the structure of the catalyst system.

[0054] In some embodiments, the catalyst system energy prediction model includes a catalyst system energy-interatomic force prediction model, and the first sample label further includes first interatomic force information corresponding to first catalyst system structure information. In some embodiments, the first interatomic force information corresponding to first catalyst system structure information can be obtained and stored by experimental measurement or quantum mechanical calculation for retrieval when needed. The first output result includes the catalyst system energy prediction result and the interatomic force information prediction result. In some embodiments, the catalyst system energy prediction result may be represented as a scalar, and the interatomic force information prediction result may be represented as a matrix.

[0055] In some embodiments, the first predetermined condition includes at least one of the following: the current first target loss is less than a predetermined threshold; and the number of iterations corresponding to the current first target loss has reached a predetermined number. For example, the predetermined number is set to 500, and the iteration stops when the number of iterations exceeds 500.

[0056] In some embodiments, step S420 includes the following steps:

[0057] First, for each first training data in the first training dataset, the first sub-loss is calculated based on the catalyst system energy prediction result corresponding to the first training data and the system energy information of the first training data. In some embodiments, the first sub-loss may be obtained by calculating the difference between the catalyst system energy prediction result corresponding to the first training data and the system energy information of the first training data.

[0058] Next, for each first training data in the first training dataset, a second sub-loss corresponding to the first training data is calculated based on the interatomic force information prediction result and the interatomic force information of the first training data. In some embodiments, the second sub-loss may be obtained by calculating the difference in each dimension between the matrix representing the interatomic force information prediction result and the matrix representing the interatomic force information and merging them as a scalar.

[0059] Finally, for each first training data point in the first training dataset, the first loss is determined based on the first sub-loss and second sub-loss corresponding to that first training data point. In some embodiments, determining the first loss based on the first sub-loss and second sub-loss may include calculating the sum of the first sub-loss and the second sub-loss.

[0060] In some embodiments, the interatomic force information of the catalytic system may be represented in the form of a three-dimensional matrix, where the data in the three dimensions represents the forces acting on the atoms in the three directions: x, y, and z. Similarly, the system structure information of the catalytic system may also be represented in the form of a three-dimensional matrix, where the data in the three dimensions represents the three-dimensional coordinates of the atoms. As can be seen, the input and output dimensions are very close, which is advantageous for training the parameters of the catalytic system energy-interatomic force prediction model.

[0061] Figure 5 shows the training of a catalyst system energy prediction model in several embodiments of this application. As shown in Figure 5, first, the 3D structure of the catalyst system (including the coordinates and numbers of each atom), and the system energy and interatomic forces corresponding to the 3D structure of the catalyst system are obtained. For example, the system energy and interatomic forces corresponding to the 3D structure of the catalyst system can be calculated based on quantum mechanics. Next, a topological map of the catalyst system is constructed based on the interatomic distances. Then, the parameters of a graph neural network are trained to obtain a catalyst system energy prediction model by taking the topological map of the catalyst system structure as input and the system energy and interatomic forces corresponding to the catalyst system structure as expected outputs.

[0062] Figure 6 is a flowchart for obtaining a catalytic system relaxation energy prediction model by training an initial prediction model of the catalytic system relaxation energy using a second training dataset. As shown in Figure 6, in some embodiments, step S350 includes steps S610, S620, S630 and S640.

[0063] In step S610, for each second training data in the second training dataset, 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.

[0064] In some embodiments, the second output result includes the catalyst system relaxation energy prediction result and the atomic displacement information prediction result. In some embodiments, the catalyst system relaxation energy prediction result is represented as a scalar, and the atomic displacement information prediction result is represented as a matrix.

[0065] In step S620, for each second training data in the second training dataset, a third loss corresponding to the second training data is calculated based on the second output result and second sample label corresponding to the second training data.

[0066] In some embodiments, the second output result includes a catalyst system relaxation energy prediction result, and step S620 includes, namely, for each second training data in the second training dataset, calculating a third loss corresponding to the second training data based on the catalyst system relaxation energy prediction result and the relaxation energy information of the second training data.

[0067] In some embodiments, a third loss may be obtained by calculating the difference between the predicted relaxation energy of the catalyst system and the relaxation energy information of the second training data.

[0068] In step S630, the second target loss of the initial prediction model of the catalyst system relaxation energy is determined based on the third loss corresponding to each second training data point in the second training dataset. In some embodiments, the second target loss may be obtained using a loss function.

[0069] In some embodiments, the second sample label further includes atomic displacement information corresponding to the second catalyst system structure information, the second output result further includes atomic displacement information prediction results, and step S630 includes, namely, for each second training data in the second training dataset, calculating a fourth loss corresponding to the second training data based on the atomic displacement information prediction results and the atomic displacement information of the second training data; and determining a second target loss of the initial prediction model of the catalyst system relaxation energy based on the third and fourth losses corresponding to each second training data in the second training dataset.

[0070] In some embodiments, the fourth loss may be obtained by calculating the differences between each dimension of the matrices and merging them as a scalar. In some embodiments, for each of the second training datasets, the sum of its third and fourth losses is calculated to obtain the overall loss, and then the sum of the overall losses corresponding to each of the second training datasets is calculated, and the result is determined as the second target loss.

[0071] In step S640, a catalyst system energy prediction model is obtained by iteratively updating the parameters of the initial prediction model of the catalyst system relaxation energy based on the second target loss until the second target loss satisfies the second predetermined condition.

[0072] In some embodiments, the initial prediction model for the catalytic system relaxation energy may include a graph neural network. Training the initial prediction model for the catalytic system relaxation energy using a second training dataset involves further training the pre-trained parameters of the graph neural network using the second training dataset, thereby obtaining a graph neural network capable of simultaneously predicting the catalytic system relaxation energy and atomic displacement based on the catalytic system structure.

[0073] In some embodiments, the second predetermined condition includes at least one of the following: namely, the current second target loss is less than a predetermined threshold; and the number of iterations corresponding to the current second target loss has reached a predetermined number. Of these, the number of iterative updates to the parameters of the initial prediction model for the catalytic system relaxation energy may be less than the number of iterative updates to the parameters of the prediction model for the catalytic system energy.

[0074] In some embodiments, the initial prediction model for the catalytic system relaxation energy may include a graph neural network, and training the initial prediction model for the catalytic system relaxation energy using a second training dataset includes, namely, further training the pre-training parameters of the graph neural network using the second training dataset to obtain a graph neural network capable of predicting the catalytic system relaxation energy based on the catalytic system structure.

[0075] In some embodiments, the first output result includes a catalyst system energy prediction result and an interatomic force information prediction result, and the catalyst system energy prediction model includes a topological 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 dataset, a topological map corresponding to the first catalyst system structure information is obtained by inputting the first catalyst system structure information corresponding to the first training data into the topological map determination submodel, and this topological map is used to represent the topological association between each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data. Subsequently, for each first training data in the first training dataset, the catalyst system energy prediction result and the interatomic force information prediction result are obtained by inputting the topological map corresponding to the first catalyst structure information corresponding to the first training data into the graph neural network submodel. In some embodiments, the graph neural network submodel includes a graph neural network, and the parameters of the graph neural network are obtained by training.

[0076] In some embodiments, the topological map includes multiple nodes and edges connecting two of each node, where each node represents an atom in the corresponding catalyst system, and each edge represents the relationship between two atoms corresponding to that edge. Obtaining the step of a topological map corresponding to the first catalyst system structure information by inputting the first catalyst system structure information corresponding to the first training data in the first training dataset into the topological map determination submodel includes the following steps: First, for each first training data in the first training dataset, the 3D coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information are determined based on the first catalyst system structure information corresponding to the first training data. In some embodiments, atoms may include adsorbent atoms, catalyst atoms, etc. Subsequently, for each first training data in the first training dataset, the position of each node in the topological map corresponding to the first catalyst system structure information and the distance between each pair of atoms are determined based on the 3D coordinates of each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data. For each pair of nodes in the topological map, the value of the edge connecting the two nodes is set to 1 if the distance between the atoms corresponding to each of the two nodes in the pair is less than or equal to a predetermined distance threshold, and the value of the edge connecting the two nodes is set to 0 if the distance between the atoms corresponding to each of the two nodes in the pair is greater than a predetermined distance threshold.

[0077] In some embodiments, obtaining the first training dataset involves the following steps: obtaining three-dimensional structural information for each of several sample catalyst systems, the three-dimensional structural information including the three-dimensional coordinates of each atom in the corresponding sample catalyst system; determining the system energy information and interatomic force information for each sample catalyst system based on the three-dimensional structural information of each sample catalyst system using a quantum mechanical method; and constructing the first training data based on the three-dimensional structural information of each sample catalyst system, the corresponding system energy information, and the interatomic force information. In some embodiments, the system energy information and interatomic force information of a catalyst system can be obtained by using quantum mechanics to calculate based on the distance between each atom in the catalyst system structure using a quantum mechanical method.

[0078] Figure 7 shows the training of a catalyst system relaxation energy prediction model in several embodiments of this application. As shown in Figure 7, first, the 3D structure of the catalyst system (including the coordinates and numbers of each atom) and the relaxation energy corresponding to the 3D structure of the catalyst system are obtained. For example, 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. Then, a topological map of the catalyst system is constructed based on the distances between atoms. Then, the parameters of a graph neural network are trained (for example, by fine-tuning the parameters based on the initial parameters) using the topological map of the catalyst system structure as input and the relaxation energy corresponding to the catalyst system structure as the expected output, thereby obtaining a catalyst relaxation energy prediction model. In some embodiments, the initial parameters of the graph neural network can be the parameters of the catalyst system energy prediction model.

[0079] Figure 8 shows the training of a multitask catalytic system relaxation energy prediction model in several embodiments of this application. As shown in Figure 8, first, the 3D structure of the catalytic system (including the coordinates and numbers of each atom), and the relaxation energy and atomic displacement corresponding to the 3D structure of the catalytic system are obtained. For example, the relaxation energy corresponding to the 3D structure of the catalytic system is obtained by calculating the difference between the system energy of the initial 3D structure of the catalytic system and the system energy of the final 3D structure. The atomic displacement corresponding to the 3D structure of the catalytic system is obtained by calculating the difference between the coordinates of each atom in the initial 3D structure of the catalytic system and the coordinates of each atom in the final 3D structure. Figure 9 shows atomic displacement in several embodiments of this application. As shown in Figure 9, the atomic displacement of a certain atom in the catalytic 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., vector R in Figure 9).

[0080] In some embodiments, the atomic coordinate information of the catalyst system is represented in the form of a three-dimensional matrix, where the data in each dimension may represent the position mapping of the atom in the three directions: x, y, and z. Similarly, the atomic displacement of the catalyst system is represented in the form of a three-dimensional matrix, where the data in each dimension may represent the position change of the atom in the three directions: x, y, and z. In conjunction with the aforementioned pre-training embodiments, it is found that the dimensions of atomic displacement and interatomic force are the same, and the dimensions of relaxation energy and system energy are the same, which is advantageous for quickly adjusting and determining appropriate parameters during training. In some embodiments, the initial parameters of the graph neural network may be the parameters of the catalyst system energy prediction model (i.e., the parameters obtained by pre-training in the aforementioned embodiments).

[0081] In some embodiments, step S320 includes the following steps: a pre-trained catalyst system energy prediction model is obtained by training a catalyst system energy prediction model using a first training dataset based on a predetermined first hyperparameter set, where the predetermined first hyperparameter set includes a predetermined first learning rate. Step S350 also includes the following steps: a catalyst system relaxation energy prediction model is obtained by training an initial catalyst system relaxation energy prediction model using a second training dataset based on a predetermined second hyperparameter set, where the second hyperparameter set includes a predetermined second learning rate, and the second learning rate is smaller than the first learning rate. In some embodiments, the second learning rate is less than or equal to one-tenth of the first learning rate. Therefore, the learning rate when training the catalyst system energy prediction model using the first training dataset is greater than the learning rate when training the initial catalyst system relaxation energy prediction model using the second training dataset. The reason for this is that the initial parameters of the latter have already been determined by the former, so only fine-tuning based on the initial parameters is required. As can be seen, the model construction method of this embodiment allows for the construction of a catalyst system relaxation energy prediction model more quickly and accurately.

[0082] In some embodiments, the first sample label further includes interatomic 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. In the pre-training and fine-tuning phases, all models being trained are multi-task predictive models, and the dimensions of their outputs are corresponding to each other (the system energy and relaxation energy data have the same dimensions, and the interatomic force and atomic displacement data have the same dimensions). Therefore, by determining the initial parameters through pre-training and then fine-tuning in the fine-tuning phase, appropriate model parameters can be quickly determined.

[0083] This application further discloses a method for predicting the relaxation energy of a catalyst system, which includes obtaining system structure information of a catalyst system awaiting prediction; and obtaining the relaxation energy of the catalyst system by inputting the system structure information of the catalyst system awaiting prediction into a catalyst system relaxation energy prediction model provided in any one embodiment of this application. In some embodiments, the catalyst system relaxation energy prediction model is run on a computing device 110 as shown in Figure 1.

[0084] Figure 10 shows a method for constructing a catalyst system relaxation energy prediction model in some embodiments of this application where multitask training is not performed. As shown in Figure 10, first, the pre-training stage is completed using a first training dataset. In the pre-training stage, first, the first training dataset is acquired, which contains 2 million data points, each containing the structure of the catalyst system and corresponding system energy information. Subsequently, a corresponding topological map is determined based on the catalyst system structure. The parameters of the graph neural network are trained by using the topological map of the catalyst system structure as input and the system energy corresponding to the catalyst system structure as the expected output, and the first parameter of the graph neural network is obtained as the result of pre-training.

[0085] Subsequently, the fine-tuning stage is completed using a second training dataset. The second training dataset contains 46 data points, each containing the initial structure of the catalyst system and its corresponding relaxation energy. Then, a corresponding topological map is determined based on the initial structure of the catalyst system. The graph neural network is trained using the topological 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 approaches the expected output, and these parameters are determined as the second parameter and used 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 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, and the graph neural network predicts the relaxation energy of the catalyst system based on the topological map of the catalyst system. The parameters of the graph neural network are the second parameter.

[0086] Figure 11 shows a method for constructing a multitask catalyst system relaxation energy prediction model in several embodiments of this application. First, a pre-training stage is completed using a first training dataset. In the pre-training stage, the first training dataset is acquired, which contains 2 million data points, each containing information on the structure of the catalyst system and the corresponding system energy and interatomic force. Subsequently, a corresponding topological map is determined based on the catalyst system structure. The first parameters of the graph neural network are obtained as a result of pre-training by training the parameters of the graph neural network with the topological map of the catalyst system structure as input and the system energy and interatomic force information corresponding to the catalyst system structure as the expected output. As can be seen, in the pre-training stage, the input to the graph neural network represents only the topological map of the system structure, but its output is the system energy and interatomic force, meaning that the graph neural network is trained to learn two tasks, and thus it belongs to multitask learning. Multitask learning is advantageous for establishing a more rigorous and comprehensive supervised function, making the process of training the graph neural network more efficient and rapid, and further increasing the adaptability of the obtained parameters.

[0087] Subsequently, the fine-tuning stage is completed using the second training dataset. The second training dataset contains 460,000 data points, each containing the initial structure of the catalyst system and its corresponding relaxation energy and atomic displacement. Then, a corresponding topological map is determined based on the initial structure of the catalyst system. The graph neural network is trained using the topological map of the initial structure of the catalyst system as input and the relaxation energy and atomic displacement 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 approaches the expected output, and these parameters are determined as the second parameter and used 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 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, and the graph neural network predicts the relaxation energy and atomic displacement 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, in the fine-tuning stage, the input to the graph neural network represents only the topological map of the system structure, but its output is relaxation energy and atomic displacement. That is, the graph neural network is trained to learn two tasks, so it still belongs to multi-task learning. Furthermore, the learning of the two tasks in the fine-tuning stage corresponds to the learning of the two tasks in the pre-training stage. This is because the dimensions of their output data are the same. As a result, by using the graph neural network parameters obtained in the pre-training stage as the initial parameters of the graph neural network in the fine-tuning stage, the optimal graph neural network parameters can be quickly determined in the fine-tuning stage, and the graph neural network obtained after fine-tuning can be made to have even higher accuracy when predicting system relaxation energy and atomic displacement.

[0088] The following table (Table I) shows examples of hyperparameters for graph neural networks during the pre-training phase.

[0089] [Table 1] Table I shows the hyperparameters used to train 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 limit the pre-training process. One “epoch” represents the process of training all samples in the training dataset once. In one epoch, the training algorithm can input all samples into the model according to the set order and perform forward propagation, loss calculation, backward propagation, and parameter updates. A batch refers to a set of samples input into the model at one time. In the neural network training process, the training data is often very large, for example, tens of thousands or even hundreds of thousands, and inputting such a large amount of data into the model at once places too much demand on the computer's performance and the neural network model's learning ability. Therefore, the training data is divided into multiple batches, and then samples from each batch are input into the model batch by batch for forward propagation, loss calculation, backward propagation, and parameter updates. "Batch size" represents the number of samples in each batch. The learning rate, also called the learning rate, represents the rate at which information is accumulated (stored) over time in a neural network. The learning rate is one of the hyperparameters that has the greatest impact on performance, and if only one hyperparameter can be adjusted, it becomes the optimal choice. Compared to other hyperparameters, the learning rate controls the effective capacity of the model in a more complex way, and the effective capacity of the model is maximized when the learning rate is optimal. Therefore, the learning rate is one of the important hyperparameters that needs to be set in order to train a neural network.

[0090] The following table (Table II) shows examples of hyperparameters for graph neural networks during the fine-tuning stage.

[0091] [Table 2] Table II shows the hyperparameters used to train the graph neural network during the fine-tuning phase in the embodiment shown in Figure 11. As shown in Table II, the number of layers in the graph neural network is 16, the learning rate is 0.00004, the batch size is 384, the epochs are 20, the relaxed energy loss factor is 1, and the delta position loss factor is 5. These hyperparameters limit the training process. Comparing Table II and Table I, we can see that the learning rate used to train the graph neural network during the fine-tuning phase is one-tenth of that during the pre-training phase. The reason for this is that during the fine-tuning phase, the initial parameters of the graph neural network are the training results from the pre-training phase, and since they are approaching the target result, only fine-tuning is necessary. Therefore, the learning rate during the fine-tuning phase can be set considerably smaller than that during the pre-training phase. As can be seen, in the fine-tuning stage, the output used to predict the predicted system energy in the "atomic-interatomic force" model is used to predict the system relaxation energy, and the output used to predict the system's interatomic forces in the "atomic-force" model is also used to predict the system's atomic displacement. Predicting the system relaxation energy is a graph-level task, and predicting the system's atomic displacement is a node-level task. Both can be trained simultaneously, making this process multi-task training. Predicting the system's positional displacement (offset) is actually an auxiliary task, and adding it can further improve the model's prediction accuracy for system relaxation energy. As can be seen, the step-built model in this embodiment allows for the rapid determination of the model's initial parameters, and then efficient and accurate model construction is achieved by fine-tuning based on those initial parameters.

[0092] The following table (Table III) shows the hyperparameters used to train the graph neural network during the pre-training phase.

[0093] [Table 3] Table III shows the hyperparameters used to train the graph neural network during the pre-training phase in the example shown in Figure 8. As shown in Table III, the graph neural network has 16 layers, a learning rate of 0.0004, a batch size of 384, and 20 epochs. These hyperparameters limit the pre-training process. As mentioned earlier, the training process of a neural network often requires dividing the training data into multiple batches. Specifically, since each batch has multiple samples, the batch size is fixed. The learning rate, also called 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 has the greatest impact on performance, and if only one hyperparameter can be adjusted, it is the best choice. Compared to other hyperparameters, the learning rate controls the effective capacity of the model in a more complex way, and the effective capacity of the model is maximized when the learning rate is optimal. Therefore, the learning rate is one of the important hyperparameters that needs to be set in order to train a neural network.

[0094] The following table (Table IV) shows the hyperparameters used to train the graph neural network during the fine-tuning phase.

[0095] [Table 4] Table IV shows the hyperparameters used to train the graph neural network during the fine-tuning phase in the embodiment shown in Figure 7. As shown in Table IV, the graph neural network has 16 layers, a learning rate of 0.00004, a batch size of 384, and 20 epochs. These hyperparameters limit the training process. Comparing Table IV with Table III, it can be seen that the learning rate used to train the graph neural network during the fine-tuning phase is one-tenth of that during the pre-training phase. The reason for this is that during the fine-tuning phase, the initial parameters of the graph neural network are the training results from the pre-training phase, and since they are close to the target result, only fine-tuning is necessary. Therefore, the learning rate during the fine-tuning phase can be set considerably smaller than that during the pre-training phase. As can be seen, the step-building model of this embodiment allows for the rapid determination of the initial parameters of the model, and then, by fine-tuning based on these initial parameters, efficient and accurate model construction can be achieved.

[0096] Figure 12 shows the effect of the catalyst system relaxation energy prediction model in several embodiments of this application. In Figure 12, the control group employs a catalyst system relaxation energy prediction model that was directly constructed without a pre-training stage, while the embodiments of this application employ a catalyst system relaxation energy prediction model that was constructed with pre-training but without multi-task training, as shown in Figure 10, and a catalyst system relaxation energy prediction model that was constructed with multi-task training, as shown in Figure 11. Four authoritative test sets were selected for the test sets, which are the 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 volumes are shown in Table V.

[0097] The following table (Table V) shows examples of authoritative test sets.

[0098] [Table 5] The catalyst relaxation energy prediction models constructed in the control group and the embodiments of this application were tested using four authoritative test sets shown in Table V, and the test results are shown in Figure 12. Prediction accuracy was evaluated by the prediction error, using the Mean Absolute Error (MAE) index, where a lower value of the MAE indicates higher prediction accuracy. In Figure 12, the black bars represent the control group, the grid bars represent the embodiment in this application where multitask training was not performed (referred to as Invention 1 of this application), and the white bars represent the embodiment in this application where multitask training was performed (referred to as Invention 2 of this application). Clearly, across these four test sets, the prediction accuracy of Invention 1 and Invention 2 of this application is higher than that of the control group. Furthermore, the figure also shows the average values ​​for four sets of values. As can be seen from the angle of the average values, Proposal 1 of this application shows a 21.6% improvement in the average prediction accuracy of the model compared to the control group, Proposal 2 of this application shows a 27.4% improvement in the average prediction accuracy of the model compared to the control group, and Proposal 2 of this application shows a 7.3% improvement in the average prediction accuracy of the model compared to Proposal 1 of this application. As can be seen, the accuracy of the models obtained by the catalytic system relaxation energy model construction method disclosed in this application is higher than that of models obtained by related technologies, and among them, the accuracy of the catalytic system relaxation energy model that has undergone multi-task training is even higher.

[0099] Figure 13 is a block diagram of an exemplary structure of a catalyst system relaxation energy prediction model construction apparatus 1300 according to several embodiments of this application.

[0100] As shown in Figure 13, the catalyst system relaxation energy prediction model construction apparatus 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.

[0101] The first acquisition module 1310 is used to acquire the first training dataset, which includes multiple first training data sets, each of which includes a first training sample and a corresponding first sample label, the first training sample including first catalyst system structure information, and the first sample label including system energy information corresponding to the first catalyst system structure information.

[0102] The first training module 1320 is used to obtain a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the first training dataset.

[0103] The first construction module 1330 is used to obtain a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the first training dataset.

[0104] The second acquisition module 1340 is used to acquire a second training dataset, which includes multiple second training data sets, each of which includes a second training sample and a corresponding second sample label, the second training sample containing second catalyst system structure information, and the second sample label containing relaxation energy information corresponding to the second catalyst system structure information.

[0105] The second training module 1350 is used to obtain a catalytic system relaxation energy prediction model by training an initial prediction model of the catalytic system relaxation energy using the second training dataset.

[0106] Therefore, the catalyst system relaxation energy prediction model construction device 1300 can construct an initial catalyst system relaxation energy prediction model based on a pre-trained catalyst system relaxation energy prediction model, thereby enabling rapid determination of the initial catalyst system relaxation energy prediction model. Subsequently, the second training module 1350 acquires a catalyst system relaxation energy prediction model by adjusting the initial catalyst system relaxation energy prediction model with the second training dataset, thereby enabling rapid acquisition of a highly accurate and fast-responding catalyst system relaxation energy prediction model. As can be seen, by using the catalyst system relaxation energy prediction model construction device 1300, a high-quality catalyst system relaxation energy prediction model can be constructed quickly and easily.

[0107] In some embodiments, the first sample label further includes interatomic force information corresponding to the structural information of the first catalyst system, and the second sample label further includes atomic displacement information corresponding to the structural information of the second catalyst system.

[0108] In some embodiments, the first training module 1320 is further used to obtain a pre-trained catalyst system energy prediction model by inputting the first training sample of each first training data in the first training dataset into the catalyst system energy prediction model; calculating the first loss corresponding to each first training data in the first training dataset based on the first output result and the first sample label of the first training data; determining the first target loss of the catalyst system energy prediction model based on the first loss corresponding to each first training data in the first training dataset; and iteratively updating the parameters of the catalyst system energy prediction model based on the first target loss until the first target loss satisfies a first predetermined condition.

[0109] In some embodiments, when the first sample label further includes interatomic force information corresponding to first catalyst system structure information, and the first output result includes catalyst system energy prediction results and interatomic force information prediction results, the first training module 1320 further calculates a first sub-loss for each first training data in the first training dataset based on the catalyst system energy prediction results and system energy information of the first training data corresponding to the first training data; calculates a second sub-loss for each first training data in the first training dataset based on the interatomic force information prediction results and interatomic force information of the first training data corresponding to the first training data; and is used to determine a first loss for each first training data in the first training dataset based on the first sub-loss and second sub-loss corresponding to the first training data.

[0110] In some embodiments, when the first sample label further includes interatomic force information corresponding to first catalyst system structure information, and the first output result includes catalyst system energy prediction results and interatomic force information prediction results, the first training module 1320 further calculates for each first training data in the first training dataset the interatomic force information prediction result corresponding to the first training data and a second loss corresponding to the first training data based on the interatomic force information of the first training data; and uses the first and second losses corresponding to each first training data in the first training dataset to determine a first target loss of the catalyst system energy prediction model.

[0111] In some embodiments, the first output results include catalyst system energy prediction results and interatomic force information prediction results, the catalyst system energy prediction model includes a topological map determination submodel and a graph neural network submodel, and the first training module 1320 further obtains a topological map corresponding to the first catalyst system structure information by inputting the first catalyst system structure information corresponding to the first training data into the topological map determination submodel for each first training data in the first training dataset, and the topological map is used to represent the topological association between each atom in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; and the topological map corresponding to the first catalyst structure information corresponding to the first training data is input to the graph neural network submodel for each first training data in the first training dataset to obtain catalyst system energy prediction results and interatomic force information prediction results.

[0112] In some embodiments, the topological map includes multiple nodes and edges connecting two of the nodes, where each node represents an atom in the corresponding catalyst system, and each edge represents the relationship between two atoms corresponding to that edge. The first training module 1320 further determines, for each first training data in the first training dataset, 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, based on the first catalyst system structure information corresponding to the first training data; and for each first training data in the first training dataset, the first training module determines, for each first training data in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data. Based on the three-dimensional coordinates of each atom, the position of each node in the topological map corresponding to the first catalyst system structure information and the distance between each pair of atoms are determined; for each pair of nodes in the topological map, the value of the edge connecting these two nodes is set to 1 if the distance between the atoms corresponding to each of the two nodes in that combination is less than or equal to a predetermined distance threshold; and for each pair of nodes in the topological map, the value of the edge connecting these two nodes is set to 0 if the distance between the atoms corresponding to each of the two nodes in that combination is greater than a predetermined distance threshold.

[0113] In some embodiments, the second training module 1350 is further used to obtain a catalyst system energy prediction model by inputting the second catalyst system structure information of the second training data into the catalyst system relaxation energy initial prediction model for each second training data in the second training dataset; calculating a third loss corresponding to the second training data based on the second output result and second sample label for each second training data in the second training dataset; determining a second target loss for the catalyst system relaxation energy initial prediction model based on the third loss corresponding to each second training data in the second training dataset; and iteratively updating the parameters of the catalyst system relaxation energy initial prediction model based on the second target loss until the second target loss satisfies a second predetermined condition.

[0114] In some embodiments, the second output result includes a catalyst system relaxation energy prediction result, and the second training module 1350 is further used to calculate a third loss corresponding to each second training data in the second training dataset, based on the catalyst system relaxation energy prediction result and the relaxation energy information of the second training data corresponding to the second training data.

[0115] In some embodiments, the second sample label further includes atomic displacement information corresponding to the second catalyst system structure information, the second output result further includes atomic displacement information prediction results, and the second training module 1350 further calculates a fourth loss corresponding to each second training data in the second training dataset based on the atomic displacement information prediction results and the atomic displacement information of the second training data; and uses the third and fourth losses corresponding to each second training data in the second training dataset to determine the second target loss of the initial prediction model of the catalyst system relaxation energy.

[0116] In some embodiments, the first acquisition module 1310 further acquires three-dimensional structural information of each of the multiple sample catalyst systems, the three-dimensional structural information of which includes the three-dimensional coordinates of each atom in the corresponding sample catalyst system; based on the three-dimensional structural information of each of the multiple sample catalyst systems, the system energy information and interatomic force information of each sample catalyst system are determined by a quantum mechanical method; and the three-dimensional structural information of each sample catalyst system, the corresponding system energy information and interatomic force information are used to construct first training data.

[0117] In some embodiments, the first training module 1320 is further used to obtain a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the first training dataset based on a predetermined first hyperparameter set, where the predetermined first hyperparameter set includes a predetermined first learning rate.

[0118] In some embodiments, the first training module 1320 is further used to obtain a catalytic system relaxation energy prediction model by training an initial prediction model of the catalytic system relaxation energy using a second training dataset based on a predetermined second hyperparameter set, wherein the second hyperparameter set includes a predetermined second learning rate, and the second learning rate is smaller than the first learning rate.

[0119] In some embodiments, the second learning rate is less than one-tenth of the first learning rate.

[0120] As shown in Figure 14, 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.

[0121] The system structure information acquisition module 1410 is used to obtain system structure information of the catalyst system awaiting prediction.

[0122] The model input module 1420 is used to obtain relaxation energy information for a catalyst system awaiting prediction by inputting system structure information of the catalyst system awaiting prediction into a catalyst system relaxation energy prediction model obtained by the catalyst system relaxation energy prediction model construction method in any one embodiment.

[0123] Figure 15 shows an exemplary system 1500, which includes an exemplary computing device 1510 representing one or more systems and / or devices capable of implementing each method in this application. The computing device 1510 includes, for example, a service provider's server, equipment connected to the server, a system-on-a-chip, and / or any other suitable computing device or computing system. The catalyst system relaxation energy prediction model construction apparatus 1300 described above with reference to Figure 9 can take the form of a computing device 1510. Alternatively, the catalyst system relaxation energy prediction model construction apparatus 1300 may be implemented as a computer program in the form of an application (app) 1516.

[0124] The illustrated illustrative computing device 1510 includes a processing system 1511, one or more computer-readable media 1512, and one or more I / O interfaces 1513, all of which are communicated with one another. Although not shown, the computing device 1510 may further include a system bus or other data and instruction transmission system used to connect various assemblies to one another. The system bus may include any one or a combination thereof of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or any one of various bus architectures, a processor or local bus. Furthermore, various other examples, such as control lines and data lines, are also possible.

[0125] The processing system 1511 represents a function that performs one or more operations using hardware. Therefore, the processing system 1511 is illustrated to include hardware elements 1514, which may be configured as processors, functional blocks, etc. These may also include other logic components formed using dedicated integrated circuits or one or more semiconductors. The hardware elements 1514 are not limited by the materials they are formed from or the processing mechanisms employed within them. For example, a processor may consist of (multiple) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In such a context, the executable instructions of the processor may also be instructions that can be executed electronically.

[0126] The computer-readable medium 1512 is illustrated to include a memory / storage device 1516. The memory / storage device 1516 represents a memory / storage capacity associated with one or more computer-readable media. The memory / storage device 1516 may include volatile media (e.g., random access memory (RAM)) and / or non-volatile media (e.g., ROM, fresh memory, optical disk, magnetic disk, etc.). The memory / storage device 1516 may also include fixed media (e.g., RAM, ROM, fixed HDD, etc.) and movable media (e.g., fresh memory, movable HDD, optical disk, etc.). The computer-readable medium 1512 may be configured in various ways as described below.

[0127] One or more I / O interfaces 1513 indicate that a user may input commands and information to the computing device 1510 using various input devices, and optionally, present information to the user and / or other assemblies or devices using various output devices. Examples of input devices include keyboards, cursor control devices (e.g., mice), microphones (e.g., for voice input), scanners, touch functions (e.g., capacitive or other sensors configured to detect physical touch), cameras (e.g., those capable of detecting gestures without touch using visible or invisible wavelengths (e.g., infrared frequencies)), etc. Examples of output devices include display units, speakers, printers, network cards, haptic response devices, etc. Thus, the computing device 1510 may be configured in various ways, as further described below, to support user interaction.

[0128] The computing device 1510 further includes application 1516. Application 1516 is, for example, a software instance for a catalyst system relaxation energy prediction model construction device 1300, which, in combination with other elements in the computing device 1510, realizes the technology described herein.

[0129] This application further provides a computer program product or computer program, which includes computer instructions, which are stored in a computer-readable storage medium. The processor of the computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, thereby causing the computing device to perform a catalytic system relaxation energy prediction model construction method provided in the various selectable implementation methods described above.

[0130] This application can describe various technologies within the general context of software elements, hardware elements, or program modules. Generally, these modules include those that perform specific tasks or implement routines, programs, objects, elements, assemblies, data structures, etc., of specific types of abstract data. As used in this application, the terms “module,” “function,” and “assembly” typically refer to software, firmware, hardware, or a combination thereof. The technical features described in this application are platform-independent, meaning that these technologies can be implemented on various computing platforms with various processors.

[0131] The modules and technical implementations described may be stored in some form of computer-readable medium or transmitted between some form of computer-readable medium. The computer-readable medium may include various media accessible to the computing device 1510. In some embodiments, the computer-readable medium may include, but is not limited to, “computer-readable storage medium” and “computer-readable signal medium”.

[0132] A “computer-readable storage medium” refers to a medium and / or device and / or tangible storage device that can store information non-temporarily, as opposed to mere signal transmission, carrier waves, or signals themselves. Therefore, a computer-readable storage medium is not a medium that carries signals. A computer-readable storage medium refers to hardware such as storage devices implemented by methods or techniques suitable for storing information (e.g., computer-readable instructions, data structures, program modules, logic elements / circuits, or other data), including volatile and non-volatile, movable and immovable media. Examples of computer-readable storage mediums may include, but are not limited to, RAM, ROM, EEPROM, fresh memory or other storage device technologies, CD-ROM, DVD or other optical storage devices, hard disks, cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or other storage devices, tangible media, or products suitable for storing information and accessible by a computer.

[0133] The term “computer-readable signal medium” refers, for example, to a signal carrier medium configured to transmit instructions to the hardware of computing device 1510 over a network. The signal medium can typically embody computer-readable instructions, data structures, program modules, or other data, for example, as a modulated data signal of a carrier, data signal, or other transmission mechanism. The signal medium may further include any information propagation medium. The term “modulated data signal” refers to a signal such that information is encoded in the signal by setting or changing one or more of the characteristics of the signal. In some embodiments, the communication medium may include, for example, a wired medium connected via a wired network, and a wireless medium such as voice, RF, infrared, and other wireless media.

[0134] As described above, the hardware element 1514 and the computer-readable medium 1512 represent instructions, modules, programmable device logic and / or fixed device logic implemented in hardware form, and in some embodiments may be used to implement at least some aspects of the technology described herein. The hardware element may include integrated circuits or systems on chips, dedicated integrated circuits (ASICs), field-programmable gate arrays (FPGAs), composite programmable logic devices (CPLDs) and assemblies of other implementations or other hardware devices in silicon. In such a context, the hardware element may be a processing device for executing program tasks defined by the instructions, modules and / or logic embodied by the hardware element, and a hardware device for storing the instructions to be executed, such as the computer-readable storage medium described above.

[0135] The aforementioned combinations can also be used to realize various technologies and modules of this application. Thus, software, hardware or program modules and other program modules can be realized as one or more instructions and / or logic embodied by a certain form of computer-readable storage medium and / or one or more hardware elements 1514. The computer 1510 may be configured to realize specific instructions and / or functions corresponding to software and / or hardware modules. Thus, for example, by using the computer-readable storage medium and / or hardware elements 1514 of the processing system, the module can be realized at least partially by hardware as a module that the computer 1510 can execute as software. The instructions and / or functions can be executed / operated by one or more products (e.g., one or more computer 1510 and / or processing systems 1511) to realize the technologies, modules and examples described in this application.

[0136] In various implementations, the computing device 1510 can adopt a variety of different configurations. For example, the computing device 1510 may be implemented as a computer device such as a personal computer, desktop computer, multi-screen computer, laptop computer, or netbook. The computing device 1510 may also be implemented as a mobile device, including, for example, a mobile phone, portable music player, portable game console, tablet computer, or multi-screen computer. The computing device 1510 may also be implemented as a television-type device, which generally has a large screen or is connected to a large screen in a casual viewing environment. These devices include televisions, set-top boxes, and game consoles.

[0137] The technology described in this application may be supported by these various configurations of the computing device 1510 and is not limited to the specific examples of the technology described in this application. The functionality can also be fully or partially implemented in a “cloud” 1520 by using a distributed system, for example, platform 1522 as described below.

[0138] Cloud 1520 includes, or represents, a platform 1522 for resource 1524. Platform 1522 abstracts the functionality of the underlying layers of hardware (e.g., servers) and software resources of Cloud 1520. Resource 1524 includes applications and / or data that can be used when performing computer processing on servers located away from computing equipment 1510. Resource 1524 may further include services provided by the Internet and / or subscriber networks, such as cellular or Wi-Fi networks.

[0139] Platform 1522 can enable computing devices 1510 to connect to other computing devices by abstracting resources and functions. Platform 1522 can further abstract the resource hierarchy to provide a corresponding level of hierarchy for the needs of resources 1524 realized via Platform 1522. Thus, in embodiments of interconnected devices, the realization of the functions described in this application can be distributed across the overall system 1500. For example, the functions may be partially realized by Platform 1522, which abstracts the functions of computing device 151 and cloud 1520.

[0140] For clarity, embodiments of this application are described by referring to different functional units. However, as is evident, without departing from this application, the function of each functional unit may be implemented in a single unit, or in multiple units or as part of other functional units. For example, a function described as being performed by a single unit may be performed by multiple different units. Therefore, references to specific functional units are merely considered to provide references to the appropriate units for the function described, and do not represent a strict logical or physical structure or configuration. Thus, this application may be implemented in a single unit, or it may be physically and functionally distributed among different units and circuits.

[0141] Furthermore, while this application is described in conjunction with several embodiments, it is not intended to be limited to any particular form described herein. Rather, the scope of this application is limited only by the attached claims. In addition, individual features may be included in different claims, but they may also be combined, and inclusion in different claims does not mean that the combination of features is impossible and / or advantageous. Moreover, the order of features in the claims does not indicate that the features must necessarily follow that order. Also, in the claims, the term "includes" does not exclude other elements, and the terms "one" or "one" do not exclude multiple elements. Reference numbers in the claims are provided as clear examples and should not be construed as limiting the scope of the claims in any way.

[0142] To ensure understanding, the specific implementation of this application includes relevant data such as software testing. When this application is applied to a specific product or technology in the form of the embodiments described above, user permission or consent will be required, and the collection, use, and processing of relevant data must comply with the relevant laws and standards of the relevant countries and regions.

[0143] The technical features of the embodiments described above can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the embodiments described above have been described. However, as long as these combinations of technical features are inconsistent, they should be considered to be included within the scope of this specification.

[0144] While preferred embodiments of this application have been described above, this application is not limited to these embodiments, and any modifications to this application that do not deviate from the spirit of this application fall within the technical scope of this application.

Claims

1. A method for constructing a catalytic system relaxation energy prediction model, which is performed by a computer, A step of obtaining a first training dataset, wherein the first training dataset includes a plurality of first training data, each first training data includes a first training sample and a corresponding first sample label, 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 step of obtaining a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the aforementioned first training dataset; A step of constructing an initial prediction model of the catalytic system relaxation energy based on the aforementioned pre-trained catalytic system energy prediction model; A step of obtaining a second training dataset, wherein the second training dataset includes a plurality of second training data, each second training data includes a second training sample and a corresponding second sample label, 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 A method comprising the step of obtaining a catalytic system relaxation energy prediction model by training the catalytic system relaxation energy prediction model using the second training dataset.

2. The method according to claim 1, A method wherein the first sample label further includes interatomic 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.

3. The method according to claim 1, The step of obtaining a pre-trained catalyst system energy prediction model by training the catalyst system energy prediction model using the aforementioned first training dataset is as follows: For each first training data in the first training dataset, 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; For each first training data in the first training dataset, the first step is to calculate the first loss corresponding to the first training data based on the first output result and the first sample label of the first training data; A step of determining the first target loss of the catalyst system energy prediction model based on the first loss corresponding to each first training data in the first training dataset; A method comprising the step of obtaining a pre-trained catalyst system energy prediction model by iteratively updating the parameters of the catalyst system energy prediction model based on the first target loss until the first target loss satisfies a first predetermined condition.

4. The method according to claim 3, The first sample label further includes interatomic force information corresponding to the first catalyst system structure information, and the first output result includes a catalyst system energy prediction result and an interatomic force information prediction result. For each first training data in the first training dataset, the step of calculating the first loss corresponding to the first training data based on the first output result and the first sample label of the first training data is: For each first training data in the first training dataset, the first sub-loss is calculated based on the catalyst system energy prediction result and the system energy information of the first training data corresponding to the first training data; For each first training data in the first training dataset, the steps of calculating a second sub-loss corresponding to the first training data based on the interatomic force information prediction result and the interatomic force information of the first training data; and A method comprising the step of determining the first loss for each first training data in the first training dataset based on a first sub-loss and a second sub-loss corresponding to the first training data.

5. The method according to claim 3, The first sample label further includes interatomic force information corresponding to the first catalyst system structure information, and the first output result includes a catalyst system energy prediction result and an interatomic force information prediction result. The step of determining the first target loss of the catalyst system energy prediction model based on the first loss corresponding to each first training data in the first training dataset is: For each first training data in the first training dataset, the steps of calculating a second loss corresponding to the first training data based on the interatomic force information prediction result and the interatomic force information of the first training data; and A method comprising the step of determining a first target loss of the catalyst system energy prediction model based on a first loss and a second loss corresponding to each first training data in the first training dataset.

6. The method according to claim 3, The first output result includes the catalyst system energy prediction result and the interatomic force information prediction result, and the catalyst system energy prediction model includes a topological map determination submodel and a graph neural network submodel. For each first training data in the first training dataset, the step of obtaining a first output result corresponding to the first training data by inputting the first training sample of the first training data into the catalyst system energy prediction model is: A step of obtaining a topological map corresponding to the first catalyst system structure information by inputting the first catalyst system structure information corresponding to the first training data in the first training dataset into the topological map determination submodel, wherein the topological map is used to represent the topological relationships between atoms in the catalyst system corresponding to the first catalyst system structure information corresponding to the first training data; and A method comprising the step of obtaining catalyst system energy prediction results and interatomic force information prediction results by inputting a topological map corresponding to the first catalyst structure information corresponding to the first training data in the first training dataset into the graph neural network submodel.

7. The method according to claim 6, The topological map includes a plurality of nodes and edges connecting each of the plurality of nodes, where each node represents an atom in the corresponding catalyst system, and each edge represents the relationship between the two atoms corresponding to that edge. For each first training data in the first training dataset, the step of obtaining a topological map corresponding to the first catalyst system structure information by inputting the first catalyst system structure information corresponding to the first training data into the topological map determination submodel is: For each first training data in the first training dataset, 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; For each first training data in the first training dataset, the step of determining the position of each node in the topological map corresponding to the first catalyst system structure information and the distance between each pair of atoms 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 pair of nodes in the topological map, the step of setting the value of the edge connecting the two nodes to 1, depending on whether the distance between the atoms corresponding to each of the two nodes in the pair is less than or equal to a predetermined distance threshold; and A method comprising the step of setting the value of the edge connecting each pair of nodes in the topological map to 0, depending on whether the distance between the atoms corresponding to each of the two nodes in the pair is greater than a predetermined distance threshold.

8. The method according to claim 1, The step of obtaining a catalyst system relaxation energy prediction model by training the catalyst system relaxation energy prediction model using the second training dataset is as follows: For each second training data in the second training dataset, the second step is to input 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 point in the second training dataset, the third loss corresponding to that second training data point is calculated based on the second output result and second sample label corresponding to that second training data point; A step of determining the second target loss of the initial prediction model of the catalyst system relaxation energy based on the third loss corresponding to each second training data in the second training dataset; and A method comprising the step of obtaining a catalytic system energy prediction model by iteratively updating the parameters of the initial prediction model of the catalytic system relaxation energy based on the second target loss until the second target loss satisfies a second predetermined condition.

9. The method according to claim 8, The second output result includes the catalyst system relaxation energy prediction result. For each second training data in the second training dataset, the step of calculating a third loss corresponding to the second training data based on the second output result and second sample label corresponding to the second training data is: A method comprising the step of calculating a third loss corresponding to each second training data in the second training dataset, based on the catalyst system relaxation energy prediction result and the relaxation energy information of the second training data.

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

11. The method according to claim 1, The step of obtaining the first training dataset is as follows: A step of obtaining three-dimensional structural information for each of several sample catalyst systems, wherein the three-dimensional structural information includes the three-dimensional coordinates of each atom in the corresponding sample catalyst system; A step of determining the system energy information and interatomic force information of each sample catalyst system by a quantum mechanical method based on the three-dimensional structural information of each sample catalyst system among the plurality of sample catalyst systems; and A method comprising the step of constructing first training data based on three-dimensional structural information of each sample catalyst system, corresponding system energy information, and interatomic force information.

12. The method according to claim 1, The step of obtaining a pre-trained catalyst system energy prediction model by training the catalyst system energy prediction model using the aforementioned first training dataset is as follows: A step of obtaining a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the first training dataset based on a predetermined first hyperparameter set, wherein the predetermined first hyperparameter set includes a predetermined first learning rate. The step of obtaining a catalyst system relaxation energy prediction model by training the catalyst system relaxation energy prediction model using the second training dataset is as follows: A method comprising the step of obtaining a catalytic system relaxation energy prediction model by training the catalytic system relaxation energy prediction model using the second training dataset based on a predetermined second hyperparameter set, wherein the predetermined second hyperparameter set includes a predetermined second learning rate, and the second learning rate is smaller than the first learning rate.

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

14. A method for predicting the relaxation energy of a catalytic system, which is performed by a computer, Steps to obtain system structure information of the catalyst system awaiting prediction; and A method comprising the step of obtaining relaxation energy information of a catalyst system awaiting prediction by inputting system structure information of the catalyst system awaiting prediction into a catalyst system relaxation energy prediction model according to any one of claims 1 to 13.

15. A device for constructing a catalytic system relaxation energy prediction model, A first acquisition module used to acquire a first training dataset, wherein the first training dataset includes a plurality of first training data, each first training data includes a first training sample and a corresponding first sample label, 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 to obtain a pre-trained catalytic system energy prediction model by training the catalytic system energy prediction model using the aforementioned first training dataset; A first construction module used to construct an initial prediction model of the catalytic system relaxation energy based on the aforementioned pre-trained catalytic system energy prediction model; A second acquisition module used to acquire a second training dataset, wherein the second training dataset includes a plurality of second training data, each second training data includes a second training sample and a corresponding second sample label, 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 Apparatus including a second training module used to obtain a catalytic system relaxation energy prediction model by training the catalytic system relaxation energy prediction model using the second training dataset.

16. It is a computing device, A memory device that stores computer programs; and Includes a processing unit connected to the memory, The processing device is configured to implement the method according to any one of claims 1 to 13 by executing the computer program.

17. A program for causing a computer to perform the method described in any one of claims 1 to 13.

Images