A method for generating a metal-organic framework for hydrogen storage in a hydrogen fuel vehicle

Abstract

The present method relates to an algorithm for generating metal-organic frameworks (MOFs) for hydrogen storage in hydrogen fuel vehicles. The MOF structure is regarded as consisting of metal clusters, organic linkers, and topological networks. The algorithm is mainly composed of Monte Carlo tree search (MCTS), which is divided into four stages: selection, expansion, simulation, and backpropagation. First, through the selection and expansion stages of MCTS, a path that can generate a complete SMILES string is obtained. Then, according to the SMILES string in the selected path, in the simulation stage, an improved GRU is used as a policy network to predict the next character of the current SMILES string, resulting in a complete SMILES string, which is the organic linker of the MOF. Subsequently, in Zeo++, the MOF is constructed using metal clusters, topological networks, and organic linkers, and then RASPA is used to simulate the hydrogen adsorption capacity of the constructed MOF. The value of this adsorption capacity is used as the reward value for the path selected by MCTS when generating the organic linker. Finally, in the backpropagation stage of MCTS, the reward value is propagated backward and the information of the nodes in the above path is updated. When the maximum value of the hydrogen adsorption capacity of the MOF remains unchanged within 7 hours, it is considered that the MOF with the best hydrogen adsorption performance has been obtained under this input.

Description

Technical Field

[0001] The present method relates to a method for generating a metal-organic framework for hydrogen storage in hydrogen fuel vehicles, which can provide theoretical guidance for the experimental synthesis of corresponding MOF materials. Background Art

[0002] Automobiles, as an indispensable means of transportation for humans, have provided convenience for human travel. However, the extensive use of fuel vehicles not only exacerbates the energy crisis problem, but also the harmful substances such as carbon oxides and sulfur oxides emitted by them will further pollute the environment. Since hydrogen is the most abundant element in the universe and hydrogen energy has the advantages of high efficiency and energy conservation, new energy vehicles powered by hydrogen have broad development prospects. Although enterprises launch hydrogen fuel vehicles with more superior performance every year, most vehicles store hydrogen by high-pressure hydrogen storage. If the hydrogen storage tank is damaged or the vehicle collides, the consequences will be unimaginable.

[0003] However, the physical adsorption hydrogen storage method requires mild conditions and fast desorption of hydrogen. Therefore, the physical adsorption hydrogen storage method has broad prospects in hydrogen fuel cell vehicles. MOF materials are a kind of nanomaterials that have received wide attention in recent years and have a three-dimensional pore structure. It is a crystalline porous material formed by the connection of metal ions or metal clusters with bridging organic ligands to form a periodic network structure. Due to their large specific surface area and high porosity, MOF materials are commonly used as adsorbents for physical adsorption in the field of gas storage. Yang et al. synthesized UiO-66 by microwave synthesis method and analyzed its hydrogen storage performance. Zhu et al. synthesized IRMOF-1 by mechanochemical method and MIL-101 by hydrothermal reaction method, and analyzed the applicability of hydrogen storage of IRMOF-1 and MIL-101 in the ship fuel cell power propulsion system. With the development of computers, computational simulation has played a very important role in the research of gas adsorption. In theory, an infinite number of MOFs can be designed using different combinations of metal clusters and organic linkers. The current deep learning makes it possible to obtain an infinite number of MOFs, and MOFs with superior performance can be designed for specific applications. Zhang et al. designed MOFs with good carbon dioxide and methane adsorption performance using MCTS. Zhang et al. designed MOFs with good carbon dioxide adsorption performance under humid conditions using MCTS. Although there has been some research on applying MOF to ship fuel cell hydrogen storage, the MOF materials used in this research have been obtained through experiments and the types of MOFs have not been increased. So far, no relevant method has been proposed to obtain MOF structures with a large hydrogen adsorption capacity based on deep learning methods and apply them to hydrogen storage in hydrogen fuel vehicles.

[0004] This method is mainly based on MCTS and is divided into four stages: selection, expansion, simulation, and backpropagation. First, through the selection and expansion stages of MCTS, a path that can generate a complete SMILES string is obtained. Then, according to the SMILES string generated by the selected path, in the simulation stage, an improved GRU is used to predict the next character of the current SMILES string. After the simulation ends, a complete SMILES string, that is, the organic linker of the MOF, will be obtained. Subsequently, in Zeo++, metal clusters, topological networks, and organic linkers are used to construct the MOF. Then, the hydrogen adsorption capacity of the constructed MOF is simulated using RASPA, and the value of this adsorption capacity is used as the reward value for the path selected by MCTS when generating the organic linker. Finally, in the backpropagation stage of MCTS, the reward value is propagated backward and the node information in the above path is updated. Summary of the Invention

[0005] This method is mainly based on MCTS and is divided into four stages: selection, expansion, simulation, and backpropagation. First, through the selection and expansion stages of MCTS, a path for generating a complete SMILES string is obtained. Then, according to the SMILES string in the selected path, in the simulation stage, an improved GRU is used as a policy network to predict the next character of the current SMILES string. After the simulation ends, a complete SMILES string, that is, the organic linker of the MOF, will be obtained. Subsequently, in Zeo++, metal clusters, topological networks, and organic linkers are used to construct the MOF. Furthermore, RASPA is used to simulate the hydrogen adsorption capacity of the constructed MOF, and the value of this adsorption capacity is used as the reward value for the path selected by MCTS when generating the organic linker. Finally, in the backpropagation stage of MCTS, the reward value is propagated backward and the information of the nodes in the above path is updated. Four steps are required to implement this method:

[0006] Step 1: Select the leaf node with the largest UCB value in the search tree;

[0007] Step 2: Add its child nodes to the leaf node obtained in Step 1 to determine the path for generating a complete SMILES string;

[0008] Step 3: Obtain a complete SMILES string, and use the organic linker, metal cluster, and topological network corresponding to this SMILES string to construct the MOF in Zeo++, and then perform hydrogen adsorption performance simulation on the constructed MOF in RASPA;

[0009] Step 4: Use the hydrogen adsorption capacity as the reward value for the path selected by MCTS when generating the organic linker. In the backpropagation stage of MCTS, the reward value is propagated backward and the information of the nodes in Step 2 is updated.

[0010] When the maximum value of the hydrogen adsorption amount of the MOF remains unchanged within 7 hours, it is considered that the MOF with the best hydrogen adsorption performance has been obtained under this input.

[0011] Compared with the prior art, the prominent features of this method are as follows:

[0012] 1. An infinite number of MOFs can be designed using different combinations of metal clusters and topological networks;

[0013] 2. The cost of chemical experiments can be reduced and the safety is high;

[0014] 3. It can provide theoretical guidance for the experimental synthesis of MOF materials with good hydrogen adsorption capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 : System block diagram of this method

[0016] Figure 2 : MOF with the experimental synthesis of metal cluster Cu2(CO2)4 and topological network rhr

[0017] Figure 3 : MOF obtained by this method using Cu2(CO2)4 metal cluster and rhr network with a hydrogen adsorption amount larger than that of the MOF in Figure 2 3-1 is the data of the input Cu2(CO2)4 metal cluster, 3-2 is the data of the input rhr topological network, and 3-3 is the MOF obtained by this method.

[0018] Figure 4 : MOF with the largest hydrogen adsorption amount obtained by this method using Cu2(CO2)4 metal cluster and topological network rhr

[0019] Figure 5 : MOF with the experimental synthesis of metal cluster Zn4O(CO2)6 and topological network pcu#1

[0020] Figure 6 : MOF obtained by this method using Zn4O(CO2)6 metal cluster and pcu#1 network with a hydrogen adsorption amount larger than that of the MOF in Figure 5 6-1 is the data of the input Zn4O(CO2)6 metal cluster, 6-2 is the data of the input pcu#1 topological network, and 6-3 is the MOF obtained by this method.

[0021] Figure 7 : MOF with the largest hydrogen adsorption amount obtained by this method using Zn4O(CO2)6 metal cluster and pcu#1 network DETAILED DESCRIPTION OF THE INVENTION

[0022] Taking the metal clusters, topological networks, and target applications (i.e., hydrogen adsorption) of MOFs as inputs, MOFs with good hydrogen adsorption performance can be obtained after being processed by this method.

[0023] The MOF obtained by using this method is called the synthesized MOF, and the experimentally synthesized MOF is called the experimental MOF. Using the Cu2(CO2)4 metal cluster and the rhr network as the first input, the comparison results between the MOF with the largest hydrogen adsorption amount obtained by using this method and the experimental MOF are shown in Table 1.

[0024] Table 1 Comparison between the MOF with the largest hydrogen adsorption amount using Cu2(CO2)4 and rhr and the experimental MOF

[0025]

[0026] Using the Zn4O(CO2)6 metal cluster and the pcu#1 network as the second input, the comparison results between the MOF with the largest hydrogen adsorption amount obtained by using this method and the experimental MOF are shown in Table 2.

[0027] Table 2 Comparison between the MOF with the largest hydrogen adsorption amount using Zn4O(CO2)6 and pcu#1 and the experimental MOF

[0028]

[0029] As can be seen from Table 1 and Table 2, this method can design MOFs with larger hydrogen adsorption amounts according to different inputs, and its structure provides theoretical guidance for the experimental synthesis of corresponding MOF materials.

Claims

1. A metal-organic framework method for hydrogen storage in hydrogen fuel vehicles, characterized in that, The method includes: Obtaining a path for generating a complete SMILES string through the selection and expansion phases of MCTS. Then, according to the SMILES string in the selected path, during the simulation phase, an improved GRU is used as a policy network to predict the next character of the current SMILES string. After the simulation ends, a complete SMILES string will be obtained, which is the organic linker of the MOF. Subsequently, in Zeo++, metal clusters, topological networks, and organic linkers are used to construct the MOF, and then RASPA is used to simulate the hydrogen adsorption capacity of the constructed MOF. The value of this adsorption capacity is used as the reward value for the path selected by MCTS when generating the organic linker. Finally, in the backpropagation phase of MCTS, the reward value is propagated backward and the information of the nodes in the path is updated. Specifically, it includes the following steps: A. Selection: Starting from the root node, the optimal child node is selected from the current nodes using specific evaluation criteria, and this process is repeated until an optimal leaf node is selected, that is, the path with the most search potential is selected. The evaluation method used here is the classical UCB function, as shown in Formula 1: Formula 1; Among them, represents the cumulative value of node i, is the cumulative access count of node i, N is the cumulative access count of the parent node of node i, C is a coefficient used to adjust the proportion of the two parts before and after the plus sign in the overall formula, and this method sets it to 1; B. Expansion: In the expansion phase, one or more nodes are added to the search tree as the children nodes of the optimal node selected in the selection phase. C. Simulation: Based on the SMILES strings in the current search tree, the trained improved GRU is used as a policy network, and this policy network is recursively used to predict the next character of the current SMILES string until the character "\n" is generated or the length of the SMILES string reaches the limit. The maximum length of the string is 81. To increase the convergence speed of GRU, the activation function adopted in this method is shown in Formula 2: Formula Two; After the simulation is completed, a complete SMILES string will be obtained. To evaluate the quality of the path selected for generating this string, that is, the size of the hydrogen adsorption capacity of the MOF constructed by the organic linker corresponding to this string, first, RDKit is used to check the validity of the SMILES string, and then it is converted into an organic molecule in 3D coordinates to obtain the atomic coordinates of the organic linker. Subsequently, the MOF is constructed in the open-source software Zeo++ by combining metal nodes, topological networks, and organic linkers. Finally, the hydrogen adsorption capacity of the MOF is evaluated by using grand canonical Monte Carlo simulation in the open-source software RASPA. In RASPA, the temperature set in this method is 298K and the pressure is 100 bar. D. Backtracking: After performing adsorption simulation on the MOF, the obtained hydrogen adsorption amount is used to evaluate the quality of the path selected for generating the SMILES string; if the organic linker corresponding to the generated SMILES string fails to combine with the metal node and the topological network to form a MOF, its value is set to 0; continuously backtrack upward until reaching the root node, and the nodes on the backtracking path will update their Q value and N value; MCTS learns how to select a better path next time according to the size of the obtained hydrogen adsorption capacity, that is, after obtaining a complete SMILES string through this path, the MOF composed of its corresponding organic linker has a greater hydrogen adsorption capacity. When the maximum value of the hydrogen adsorption capacity of the MOF remains unchanged within 7 hours, it is considered that the MOF with the best hydrogen adsorption performance has been obtained under the input of the SMILES string, and the algorithm terminates and is completed.

Images