An organic-inorganic hybrid perovskite performance prediction method, prediction system and device
By using quantum graph neural networks to process the framework structure information of organic-inorganic hybrid perovskites, the problem of insufficient research on structure-property relationships in existing technologies is solved, and efficient material property prediction and design are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MINJIANG UNIVERSITY
- Filing Date
- 2023-07-06
- Publication Date
- 2026-04-14
AI Technical Summary
Existing technologies lack diverse methods for predicting the properties of organic-inorganic hybrid perovskite materials, resulting in insufficient research on structure-property relationships and a large number of unexplored potential materials in materials databases.
By combining quantum graph neural networks with classical quantum collaborative computing systems, organic and inorganic framework structural information is extracted, encoded into quantum states, and aggregated. The structure-property relationship is then predicted using predictive quantum graph neural networks.
It enables efficient prediction of structure-property relationships of organic-inorganic hybrid perovskite materials on a quantum computing platform, provides a fully automated data processing workflow, shortens the material design cycle, and reduces R&D costs.
Smart Images
Figure CN116796834B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of quantum neural network technology, and in particular to a method, system and device for predicting the properties of organic-inorganic hybrid perovskites. Background Technology
[0002] Organic-inorganic perovskite materials are a large class of materials with an ABX3 crystal structure and have wide applications in photovoltaic devices. Current performance studies on organic-inorganic perovskite materials focus on framework structure, band gap, and total energy. However, due to the limited methods for predicting these structures and properties, the structure-property relationships of a large number of organic-inorganic hybrid perovskites remain to be studied in materials databases, reported literature, and the potential synthetic material space. Summary of the Invention
[0003] Purpose of the invention: This application provides a method, system and device for predicting the properties of organic-inorganic hybrid perovskites, which can solve the structure-property relationship of organic-inorganic hybrid perovskites using quantum graph neural networks under the quantum computing platform and the classical quantum cooperative computing system architecture.
[0004] This application discloses a method for predicting the properties of organic-inorganic hybrid perovskites, the specific steps of which are as follows:
[0005] S1 takes the structural information of any organic-inorganic hybrid perovskite as input and marks it as structural data S0;
[0006] S2, extract organic skeleton structure information from structural data S0, input it into the first quantum graph neural network used to extract organic skeleton structure features, encode the organic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects, obtain the evolved quantum state information to optimize the first quantum graph neural network;
[0007] S3: Extract inorganic skeleton structure information from structural data S0 and input it into the second quantum graph neural network used to extract inorganic skeleton structure features. After encoding the inorganic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network.
[0008] S4. The feature descriptions of organic and inorganic skeleton structure information are aggregated and encoded into quantum state data. The aggregated quantum state data is then input into a predictive quantum graph neural network with learning capabilities, and the predictive quantum graph neural network is trained and optimized.
[0009] S5. The optimized first quantum graph neural network and the optimized second quantum graph neural network are used to obtain the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite and aggregate them. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical properties and / or chemical properties of the target organic-inorganic hybrid perovskite.
[0010] Preferably, in step S2, the specific steps for extracting organic framework structural information from structural data S0 and encoding it into quantum state data are as follows:
[0011] Based on the organic skeletal structure information SA, the adjacency matrix DA1 corresponding to the organic molecular structure is calculated. The atomic vectors of the organic molecular skeletal structure are calculated by combining SA and DA1. The order of the vectors of different atoms is matched with the molecular topology reflected in DA1 to obtain the organic atomic vector data DA2. The result of combining DA1 and DA2 is encoded as an input quantum state and then fed into the first quantum graph neural network for calculation.
[0012] Preferably, in step S2, the specific steps for optimizing the first quantum graph neural network are as follows:
[0013] The result DA3, which combines DA1 and DA2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement result. The parameterized quantum efficiency updates the weights in the first quantum graph neural network and updates the node and edge information DA4 of the graph network.
[0014] Preferably, in step S3, the specific steps for extracting inorganic framework structural information from structural data S0 and encoding it into quantum state data are as follows:
[0015] The adjacency matrix DB1 of the inorganic framework is calculated based on the topological structure in the inorganic framework structure information SB, and the inorganic atom vector data DB2 is calculated based on the atomic structure information in the inorganic framework. The result of combining DB1 and DB2 is encoded as an input quantum state and then fed into the second quantum graph neural network for calculation.
[0016] Preferably, in step S3, the specific steps for optimizing the second quantum graph neural network are as follows:
[0017] The result DB3, which combines DB1 and DB2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement results. The parameterized quantum efficiency updates the weights of the second quantum graph neural network and updates the node and edge information DB4 of the graph network.
[0018] Preferably, the adjacency matrix describes the atomic skeleton information in the molecule, including atomic skeleton information and charge virtual bits, wherein the charge virtual bits are used to describe the molecule's charge information.
[0019] Preferably, in step S4, the aggregation methods for organic skeleton structural features and inorganic skeleton structural features include vector concatenation, matrix concatenation, and tensor concatenation.
[0020] Preferably, in step S4, the optimization steps of the predictive quantum graph neural network with learning function are as follows:
[0021] The aggregated data is input into a predictive quantum graph neural network with learning capabilities to obtain the output result DC1. Combining DC1 and the data information of the labels, the loss function of supervised learning is calculated, and a classical optimization algorithm is selected to update the entire network structure, thus completing the learning and training of the predictive quantum graph neural network. The labels include the physical and / or chemical properties of organic-inorganic hybrid perovskites.
[0022] Furthermore, when using structural information files of organic-inorganic hybrid perovskites as input, including but not limited to structural input files such as cif, POSCAR, and xyz that can be converted to each other using existing software tools.
[0023] Furthermore, when organic framework structure information is input into the first quantum graph neural network or inorganic framework structure information is input into the second quantum graph neural network for data processing, the structure information is classical data. When calling the quantum neural network (QNN) with quantum computing capabilities, the classical data needs to be encoded into quantum states and input into the QNN. At this time, the QNN replaces the multilayer perceptron in the ordinary message passing network (MPN). The evolution of quantum state data in the quantum circuit corresponding to the quantum neural network (QNN) corresponds to the description of quantum physics and quantum information. The information in the evolved quantum state is obtained through quantum measurement. The classical data obtained by measurement is used to update the first or second quantum graph neural network. The quantum graph neural network satisfies both the deep learning characteristics of graph neural networks and the characteristics of quantum computing.
[0024] Furthermore, quantum graph neural networks (GNNs) simultaneously incorporate features of quantum computing and deep learning. The quantum computing characteristics of GNNs are manifested in the construction, evolution, and measurement of quantum states; the deep learning characteristics are reflected in the fact that the control variables for the evolutionary process are classical information, allowing for learning and updating entirely in the manner of a deep learning optimizer. For example, deep learning can be performed on existing supervised learning tasks, where the phase information of the Pauli rotation gate within the GNN is obtained through quantum measurement and used as classical variable data for updating the GNN.
[0025] Furthermore, in the process of deep learning, the construction of the loss function is related to the labels and the predicted output values, both of which are classic data.
[0026] Furthermore, when calculating the adjacency matrix of the organic and inorganic frameworks, the calculation can be completed using the algorithm flow for calculating the molecular adjacency matrix from the molecular structure provided by chemical information tools such as RDKit. When calculating the atomic vectors of the organic and inorganic frameworks, the components described by the vector of each atom contain data reflecting the atomic charge distribution, atomic element type, atomic spatial position, and other physicochemical properties.
[0027] Furthermore, organic-inorganic hybrid perovskites are typically ionic crystals, with the organic framework molecules being charged or exhibiting strong polarization. By adding a virtual charge site to the graph network where atoms act as nodes, describing the charge and its location, we can better reflect the microscopic effects of organic-inorganic hybrid perovskite materials. These microscopic effects are also objectively related to the macroscopic properties of the material. Therefore, adding virtual charge sites can more accurately describe the structure of organic-inorganic hybrid perovskites.
[0028] Furthermore, the target physical and / or chemical properties of the organic-inorganic hybrid perovskite solved by the predictive quantum graph neural network are the same as the labeled data in the learning task. In the supervised learning task, the labeled data are the properties of different configurations. The data that can be used as labels include physical and chemical properties such as binding energy, electronic band gap, and optical absorption coefficient.
[0029] Furthermore, based on the unified interface for graph neural network input data, graph neural networks can employ, but are not limited to, graph neural network algorithms running under message-passing networks such as GCN graph convolutional neural networks, GIN graph isomorphic neural networks, and GAT graph attention mechanism neural networks. They can also handle data-driven learning tasks involving organic and inorganic perovskites according to the unified software architecture of quantum computing. The unified software architecture for quantum computing can be referenced in the definition of "quantum data, classical control" in *Fundamentals of Quantum Programming*, edited by Ying Mingsheng.
[0030] The beneficial effects of this application are as follows: First, a quantum graph neural network is obtained by combining quantum computers with graph neural network deep learning tools to process the prediction of properties such as band gap and total energy of organic-inorganic hybrid perovskite materials; Second, it can provide processing methods for the structural information of organic and inorganic frameworks, so that they can meet the input data requirements of quantum graph neural networks, and can use graph neural networks to realize the feature description of framework information; Third, it can realize the aggregation of inorganic and organic framework information, and use the aggregated data to predict further processing of quantum graph neural networks to achieve the purpose of structure-property relationship prediction; Fourth, in the process of structure-property relationship prediction, it can realize a fully automatic data processing flow and complete the data-driven quantum neural network training and learning process. Attached Figure Description
[0031] Figure 1 This is a flowchart of the workflow of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural networks of the present invention;
[0032] Figure 2 This is the structural formula of the organic framework in a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural network of the present invention;
[0033] Figure 3 This is a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural network of the present invention, which describes the edge index information in the graph neural network of organic framework structure;
[0034] Figure 4 This is the adjacency matrix corresponding to the inorganic skeleton topology graph network in a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural network of the present invention;
[0035] Figure 5 This is the graph structure data of the inorganic framework in a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural network of the present invention;
[0036] Figure 6 This is the output result of processing inorganic skeleton graph structure data using a classical graph neural network in a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural networks of the present invention;
[0037] Figure 7 This is the output result of processing inorganic skeleton graph structure data using a quantum graph neural network in a specific embodiment of the organic-inorganic hybrid perovskite prediction method based on quantum graph neural networks of the present invention;
[0038] Figure 8 This is a schematic diagram of the structure of the organic-inorganic hybrid perovskite performance prediction system based on quantum graph neural network of the present invention. Detailed Implementation
[0039] The technical solutions in this application will be clearly and completely described below with reference to the embodiments of this application.
[0040] Perovskite materials are a large class of materials with an ABX3 crystal structure. Typically, X is an anion, A and B are cations, and BX forms a cubic octahedron centered on B, with A interspersed in the octahedron. In organic-inorganic hybrid perovskites, which are representative in photovoltaic devices, A is the organic molecular framework, B is a metal element such as Pb, Sn, or Ge, and X is a halide anion. MAPbI3 was the first material used in perovskite photovoltaic devices, and FAPbI3 and FAPbBr3 are also important materials for fabricating perovskite devices. Deep learning tools such as graph neural networks for predicting perovskite properties have been widely accepted and are considered helpful in promoting and accelerating material design, shortening development cycles, and reducing costs. For example, a research team at KAIST in South Korea collaborated with Professor Aron Walsh's team at Imperial College London to use graph neural networks to accelerate the synthesis and design of perovskites. This work was published in 2022 in NPJ Computational Materials, a Nature collaboration journal. These innovative achievements demonstrate the cutting-edge and innovative nature of using deep learning and other methods in perovskite research both domestically and internationally.
[0041] The fundamental logic of graph neural networks (GNNs), a deep learning tool, relies on message passing networks (MPNs). During the training and learning process of MPNs, updates to the graph network's topology, edges, and node data are required. As the number of nodes and the complexity of the topology increase, the computational load during GNN training increases rapidly. When GNNs are used to predict crystal structures such as perovskites, each node typically corresponds to an atom vector. However, the number of atoms involved in actual materials research is extremely large, potentially representing a huge demand for intelligent computing power. Quantum computers are considered a highly efficient and powerful tool. IBM, a leading international quantum computing giant, is continuously improving the size of qubits and quantum volume to enhance computing power. IBM has also constructed and run quantum neural networks on its quantum computing platform, enabling quantum computers to bring improvements in computing power and other algorithmic logic to deep learning models.
[0042] There are still many structure-property relationships (SPRs) of organic-inorganic hybrid perovskites that remain to be studied in materials databases, reported literature, and the potential synthetic material space. Therefore, using neural networks on a quantum platform to solve the SPRs of organic-inorganic hybrid perovskites has broad research and application prospects.
[0043] A method for predicting the properties of organic-inorganic hybrid perovskites, the specific steps of which are as follows:
[0044] S1 takes the structural information of any organic-inorganic hybrid perovskite as input and marks it as structural data S0;
[0045] S2, extract organic skeleton structure information from structural data S0, input it into the first quantum graph neural network used to extract organic skeleton structure features, encode the organic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects, obtain the evolved quantum state information to optimize the first quantum graph neural network;
[0046] S3: Extract inorganic skeleton structure information from structural data S0 and input it into the second quantum graph neural network used to extract inorganic skeleton structure features. After encoding the inorganic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network.
[0047] S4. The feature descriptions of organic and inorganic skeleton structure information are aggregated and encoded into quantum state data. The aggregated quantum state data is then input into a predictive quantum graph neural network with learning capabilities, and the predictive quantum graph neural network is trained and optimized.
[0048] S5. The optimized first quantum graph neural network and the optimized second quantum graph neural network are used to obtain the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite and aggregate them. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical properties and / or chemical properties of the target organic-inorganic hybrid perovskite.
[0049] Understandably, in the context of studying the structure-property relationship of organic-inorganic hybrid perovskites, this invention proposes a quantum graph neural network (QNN) that can be used for the design of this type of material, along with a complete process for building a model for this scenario from scratch. This includes: (1) a scheme for extracting the spatial coordinates of organic atoms from standard structure files (cif or POSCAR) and converting them into graph network input data; (2) a scheme for constructing the quantum graph neural network input from the inorganic skeleton in the structure file; (3) training and learning for embedding / feature extraction of the quantum graph neural network for both the organic and inorganic skeletons; (4) a quantum neural network that fuses the organic and inorganic skeleton information; and (5) a training and optimization scheme for the entire quantum graph neural network. This invention can effectively address the gap in solving the structure-property relationship of organic-inorganic hybrid perovskites using quantum graph neural networks within a quantum computing platform and a classical quantum-coordinated computing architecture.
[0050] This embodiment provides a classical computing-quantum computing cross-execution scheme that can run on environments such as quantum computing platforms, classical computing platforms, and quantum simulators, and can be used for data-driven computing tasks on organic-inorganic hybrid perovskites, with graph neural networks as the main deep learning backbone.
[0051] Based on the unified interface for graph neural network input data, this embodiment supports, but is not limited to, graph neural network algorithms running under message passing networks, such as GCN graph convolutional neural network, GIN graph isomorphic neural network, and GAT graph attention mechanism neural network, and processes data-driven learning tasks of organic and inorganic perovskites according to the unified software architecture of quantum computing.
[0052] like Figure 1 As shown, the specific steps are as follows:
[0053] Step 1: The structural information of EAGeF3 organic-inorganic hybrid perovskite is used as input and labeled as structural data S0;
[0054] Step 2: Extract organic framework structural information EA from structural data S0. + EA is an Ethylamine molecule with the structural formula shown below. Figure 2 As shown, the graph structure data in the first quantum graph neural network with a message-passing mechanism includes two parts: node vector x and edge index edge_index. Based on the organic skeleton structure information EA... + This paper presents an algorithm for calculating the molecular adjacency matrix from molecular structures using chemical information tools such as RDKit. The adjacency matrix DA1 corresponding to the organic molecular structure is calculated as the edge index data, such as... Figure 3 As shown, in this embodiment, the adjacency matrix, in its description of the atomic skeleton information in the molecule, introduces a virtual bit to describe the molecule's charge information. DA1 is constructed by combining the atomic skeleton information and the charge virtual bit. This is then combined with the organic skeleton structure information EA. + Using the adjacency matrix DA1, the atomic vector DA2 of the organic molecular skeleton is calculated. In this embodiment, the atomic vector DA2 includes attributes such as atomic number, spatial coordinates, and physicochemical properties, and adds a description of charge positions, such as... Figure 2 As shown, corresponding to the molecular topology reflected in DA1, positions 0, 1, and 2 correspond to C, C, and N atoms respectively, and position 3 corresponds to the charge site, which is used to describe the charge information. The connection relationship can be derived from the adjacency matrix to reflect the charge information of different atoms.
[0055] The result DA3, which combines DA1 and DA2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement result. The parameterized quantum efficiency updates the weights in the first quantum graph neural network and updates the node and edge information DA4 of the graph network.
[0056] Step 3: Extract the inorganic skeleton structure information GeF3 from the structural data S0, and calculate the adjacency matrix DB1 of the inorganic skeleton based on the topological structure of the inorganic skeleton structure information GeF3, such as... Figure 4 As shown, inorganic atom vector data DB2 is calculated based on the atomic structure information in the inorganic framework. The result of combining DB1 and DB2 is encoded as an input quantum state and then fed into the second quantum graph neural network for calculation. In this embodiment, a non-periodic fractional graph network relationship is used, where the fractional proportion is reflected in the description of the atom vectors, describing all atoms in a clockwise direction and passing through the central atom. The encoded graph structure data of GeF3 is as follows. Figure 5 As shown.
[0057] The encoded GeF3 graph structure data is then processed by a quantum neural network composed of parameterized quantum circuits to obtain measurement results. The parameterized quantum efficiency is used to update the weights in the second quantum graph neural network, and the node and edge information DB4 of the graph network is also updated. For example... Figure 7 The image shows the output of processing inorganic skeleton graph structure data using a second quantum graph neural network. Alternatively, a classical graph neural network can be used to process the input of GeF3, yielding the following output: Figure 6 As shown.
[0058] Step 4: The feature descriptions of the organic and inorganic framework structures are aggregated and encoded into quantum state data. The aggregated quantum state data is then input into a predictive quantum graph neural network with learning capabilities, and the predictive quantum graph neural network is trained and optimized. In this embodiment, the aggregation methods for the feature descriptions of the organic and inorganic framework structures include, but are not limited to, vector concatenation, matrix concatenation, and tensor concatenation. The steps for training and optimizing the predictive quantum graph neural network include: inputting the aggregated data into the predictive quantum graph neural network with learning capabilities to obtain the output result DC1; combining DC1 and the label data information to calculate the supervised learning loss function; and selecting a classical optimization algorithm to update the entire network structure, thus completing the learning and training of the predictive quantum graph neural network. In this embodiment, the label information consists of the physical and chemical properties of the organic-inorganic hybrid perovskite, including but not limited to physical and chemical properties such as electronic band gap and optical absorption coefficient.
[0059] Step 5: Using the optimized first and second quantum graph neural networks, the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite are obtained and aggregated. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical and / or chemical properties of the target organic-inorganic hybrid perovskite.
[0060] In summary, this technical solution, in the context of studying the structure-property relationship of organic-inorganic hybrid perovskites, extracts organic and inorganic framework structural information from standard structural files, converts it into graph network input data, and uses graph neural networks for processing to obtain the evolution results of organic and inorganic framework structural information. Finally, through the aggregation of inorganic and organic framework information, it is used to achieve optimized training and structure-property prediction of predictive quantum graph neural networks, effectively filling the gap in solving the structure-property relationship of organic-inorganic hybrid perovskites using quantum graph neural networks under the computing architecture of quantum computing platforms and classical quantum cooperative systems.
[0061] Based on the same concept, embodiments of this application also provide an organic-inorganic hybrid perovskite performance prediction system, such as... Figure 8 As shown, it includes:
[0062] The input module is configured to take the structural information of any organic-inorganic hybrid perovskite as input, and label it as structural data S0;
[0063] The first training module is configured to extract organic skeleton structure information from structural data S0, input it into a first quantum graph neural network for extracting organic skeleton structure features, encode the organic skeleton structure information into quantum states, call a quantum neural network with quantum computing special effects, and obtain the evolved quantum state information for optimizing the first quantum graph neural network.
[0064] The second training module is configured to extract inorganic skeleton structure information from the structural data S0, input it into the second quantum graph neural network used to extract inorganic skeleton structure features, encode the inorganic skeleton structure information into quantum states, and then call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network;
[0065] The third training module is configured to aggregate the feature descriptions of organic and inorganic framework structure information and encode them into quantum state data, input the aggregated quantum state data into a predictive quantum graph neural network with learning function, and train and optimize the predictive quantum graph neural network.
[0066] The output module uses optimized first and second quantum graph neural networks to obtain and aggregate the organic and inorganic framework structural features of the target organic-inorganic hybrid perovskite. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical and / or chemical properties of the target organic-inorganic hybrid perovskite.
[0067] It should be noted that the organic-inorganic hybrid perovskite performance prediction system of this embodiment corresponds to the aforementioned organic-inorganic hybrid perovskite performance prediction method. The functional modules in the system correspond to the respective steps in the prediction method. The organic-inorganic hybrid perovskite performance prediction system of this embodiment can be implemented in conjunction with the aforementioned organic-inorganic hybrid perovskite performance prediction method. Accordingly, the relevant technical details mentioned in the organic-inorganic hybrid perovskite performance prediction system of this embodiment can also be applied to the organic-inorganic hybrid perovskite performance prediction method. Furthermore, the aforementioned functional modules can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. Additionally, these modules can be fully or partially integrated together, or implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, some or all steps of the aforementioned method, or the aforementioned functional modules, can be completed through hardware integrated logic circuits in the processor element or through software instructions.
[0068] The embodiment also provides an organic-inorganic hybrid perovskite performance prediction device, including a processor, a memory, and a communication interface connected to a system bus. The processor provides control and computational capabilities; the memory stores a computer program, which, when executed by the processor, implements the organic-inorganic hybrid perovskite performance prediction method. The memory includes a computer storage medium and internal memory. The computer storage medium is a non-volatile storage medium storing an operating system and the computer program, while the internal memory provides an environment for the operation of the operating system and the computer program. The communication interface of this computer device is used for wired or wireless communication with external terminals, such as via Wi-Fi or mobile cellular networks.
[0069] In some embodiments, when a processor executes computer program instructions, it performs the following steps:
[0070] S1 takes the structural information of any organic-inorganic hybrid perovskite as input and marks it as structural data S0;
[0071] S2, extract organic skeleton structure information from structural data S0, input it into the first quantum graph neural network used to extract organic skeleton structure features, encode the organic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects, obtain the evolved quantum state information to optimize the first quantum graph neural network;
[0072] S3: Extract inorganic skeleton structure information from structural data S0 and input it into the second quantum graph neural network used to extract inorganic skeleton structure features. After encoding the inorganic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network.
[0073] S4. The feature descriptions of organic and inorganic skeleton structure information are aggregated and encoded into quantum state data. The aggregated quantum state data is then input into a predictive quantum graph neural network with learning capabilities, and the predictive quantum graph neural network is trained and optimized.
[0074] S5. The optimized first quantum graph neural network and the optimized second quantum graph neural network are used to obtain the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite and aggregate them. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical properties and / or chemical properties of the target organic-inorganic hybrid perovskite.
[0075] The above provides a detailed description of the method, system, and apparatus for predicting the performance of organic-inorganic hybrid perovskites provided in the embodiments of this application. Specific examples have been used in this application to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the technical solutions and core ideas of this application. Those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method for predicting the properties of organic-inorganic hybrid perovskites, characterized in that, The specific steps are as follows: S1 takes the structural information of any organic-inorganic hybrid perovskite as input and marks it as structural data S0; S2, extract organic skeleton structure information from structural data S0, input it into the first quantum graph neural network used to extract organic skeleton structure features, encode the organic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects, obtain the evolved quantum state information to optimize the first quantum graph neural network; S3: Extract inorganic skeleton structure information from structural data S0 and input it into the second quantum graph neural network used to extract inorganic skeleton structure features. After encoding the inorganic skeleton structure information into quantum states, call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network. S4. The feature descriptions of organic and inorganic skeleton structure information are aggregated and encoded into quantum state data. The aggregated quantum state data is then input into a predictive quantum graph neural network with learning capabilities, and the predictive quantum graph neural network is trained and optimized. S5. The optimized first quantum graph neural network and the optimized second quantum graph neural network are used to obtain the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite and aggregate them. The aggregated data is encoded into quantum state inputs to the optimized predictive quantum graph neural network to solve for the target physical properties and / or chemical properties of the target organic-inorganic hybrid perovskite. In step S2, the specific steps for extracting organic framework structural information from structural data S0 and encoding it into quantum state data are as follows: The adjacency matrix DA1 corresponding to the organic molecular structure is calculated based on the organic skeleton structure information SA. The atomic vectors of the organic molecular skeleton are calculated by combining SA and DA1. The order of the vectors of different atoms is matched with the molecular topology reflected in DA1 to obtain the organic atomic vector data DA2. The result of combining DA1 and DA2 is encoded as an input quantum state and then fed into the first quantum graph neural network for calculation. In step S2, the specific steps for optimizing the first quantum graph neural network are as follows: The result DA3, which combines DA1 and DA2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement result. The parameterized quantum efficiency updates the weights in the first quantum graph neural network and updates the node and edge information DA4 of the graph network. In step S3, the specific steps for extracting inorganic framework structural information from structural data S0 and encoding it into quantum state data are as follows: The adjacency matrix DB1 of the inorganic framework is calculated based on the topological structure in the inorganic framework structure information SB, and the inorganic atom vector data DB2 is calculated based on the atomic structure information in the inorganic framework. The result of combining DB1 and DB2 is encoded as an input quantum state and then fed into the second quantum graph neural network for calculation. In step S3, the specific steps for optimizing the second quantum graph neural network are as follows: The result DB3, which combines DB1 and DB2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement results. The parameterized quantum efficiency updates the weights of the second quantum graph neural network and updates the node and edge information DB4 of the graph network. The adjacency matrix described in the molecule includes atomic skeleton information and charge virtual bits, which are used to describe the molecule's charge information.
2. The method for predicting the properties of organic-inorganic hybrid perovskites according to claim 1, characterized in that: In step S4, the aggregation methods for organic and inorganic skeleton structural feature data include vector concatenation, matrix concatenation, and tensor concatenation.
3. The method for predicting the properties of organic-inorganic hybrid perovskites according to claim 2, characterized in that: In step S4, the optimization steps of the predictive quantum graph neural network with learning capabilities are as follows: The aggregated data is input into a predictive quantum graph neural network with learning capabilities to obtain the output result DC1. Combining DC1 and the data information of the labels, the loss function of supervised learning is calculated, and a classical optimization algorithm is selected to update the entire network structure, thus completing the learning and training of the predictive quantum graph neural network. The labels include the physical and / or chemical properties of organic-inorganic hybrid perovskites.
4. A performance prediction system for organic-inorganic hybrid perovskites, characterized in that, include: An input module is configured to take the structural information of any organic-inorganic hybrid perovskite as input, and label it as structural data S0; The first training module is configured to extract organic skeleton structure information from structural data S0, input it into a first quantum graph neural network for extracting organic skeleton structure features, encode the organic skeleton structure information into quantum states, call a quantum neural network with quantum computing special effects, and obtain the evolved quantum state information for optimizing the first quantum graph neural network. The second training module is configured to extract inorganic skeleton structure information from the structural data S0, input it into the second quantum graph neural network for extracting inorganic skeleton structure features, encode the inorganic skeleton structure information into quantum states, and then call the quantum neural network with quantum computing special effects to obtain the evolved quantum state information for optimizing the second quantum graph neural network. The third training module is configured to aggregate the feature descriptions of organic and inorganic framework structure information and encode them into quantum state data, input the aggregated quantum state data into a predictive quantum graph neural network with learning function, and train and optimize the predictive quantum graph neural network. The output module uses the optimized first quantum graph neural network and the second quantum graph neural network to obtain the organic framework structure features and inorganic framework structure features of the target organic-inorganic hybrid perovskite and aggregates them. The aggregated data is encoded into quantum states and input into the optimized predictive quantum graph neural network to solve for the target physical properties and / or chemical properties of the target organic-inorganic hybrid perovskite. In the first training module, the specific steps for extracting organic framework structural information from structural data S0 and encoding it into quantum state data are as follows: The adjacency matrix DA1 corresponding to the organic molecular structure is calculated based on the organic skeleton structure information SA. The atomic vectors of the organic molecular skeleton are calculated by combining SA and DA1. The order of the vectors of different atoms is matched with the molecular topology reflected in DA1 to obtain the organic atomic vector data DA2. The result of combining DA1 and DA2 is encoded as an input quantum state and then fed into the first quantum graph neural network for calculation. The specific steps for optimizing the first quantum graph neural network in the first training module are as follows: The result DA3, which combines DA1 and DA2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement result. The parameterized quantum efficiency updates the weights in the first quantum graph neural network and updates the node and edge information DA4 of the graph network. In the second training module, the specific steps for extracting inorganic framework structural information from structural data S0 and encoding it into quantum state data are as follows: The adjacency matrix DB1 of the inorganic framework is calculated based on the topological structure in the inorganic framework structure information SB, and the inorganic atom vector data DB2 is calculated based on the atomic structure information in the inorganic framework. The result of combining DB1 and DB2 is encoded as an input quantum state and then fed into the second quantum graph neural network for calculation. The specific steps for optimizing the second quantum graph neural network in the second training module are as follows: The result DB3, which combines DB1 and DB2, is encoded according to the mathematical characteristics of the quantum state density matrix. The encoded result is then processed by a quantum neural network composed of parameterized quantum circuits to obtain the measurement results. The parameterized quantum efficiency updates the weights of the second quantum graph neural network and updates the node and edge information DB4 of the graph network. The adjacency matrix described in the molecule includes atomic skeleton information and charge virtual bits, which are used to describe the molecule's charge information.
5. A device for predicting the performance of organic-inorganic hybrid perovskites, characterized in that, This includes the processor and the memory that stores computer program instructions; When the processor executes computer program instructions, it implements the organic-inorganic hybrid perovskite prediction method according to any one of claims 1 to 3.