Systematic Study Of MOFs For Hydrogen Storage In Hydrogen Fuel Cell Vehicles Based On MCTS And Improved GR

Hydrogen fuel vehicles can achieve zero-pollution emissions and are considered to be the mainstream of the future automotive industry.Although the research of hydrogen fuel vehicles has made great progress,the hydrogen storage system has always been the main technical obstacle hindering its development.At present,hydrogen fuel vehicles mainly use high-pressure hydrogen storage technology,but this technology has high requirements for the material of the gas cylinder,and there are some potential hazards.Therefore,there is an urgent need to develop a safe and efficient hydrogen storage technology.Metal Organic Frameworks（MOFs）are formed by connecting metal clusters and organic linkers according to different topologies,and are considered as the most potential candidate hydrogen storage materials.Because it takes a lot of cost to prepare and test the properties of MOFs experimentally,it is more promising to obtain MOFs with good hydrogen adsorption performance through reversing design through machine learning.Therefore,this paper proposes a system based on the reverse design of Monte Carlo Tree Search（MCTS）and improved Gate Recurrent Unit（GRU）to generate MOFs for hydrogen storage in hydrogen fuel vehicles.The specific methods are as follows:We represent the organic linker in MOFs as a SMILES string,and each node in the MCTS search tree corresponds to a SMILES character.In order to improve the prediction accuracy of the strategy network in MCTS for the next character of the current SMILES string,the tanh function in the GRU neuron is improved,and the improved GRU layer is used as the strategy network in MCTS.Recursively use the strategy network in MCTS to get the complete SMILES string,then use metal cluster,topology and organic linker to generate MOFs,and perform hydrogen adsorption simulations on them.Finally,the simulated hydrogen adsorption amount is used as the reward value to propagate back in the backpropagation of MCTS,and the information of the nodes in the search tree is updated,so as to guide the next search towards the direction that can generate the best organic linker.The experimental results show that the prediction accuracy of the method proposed in this paper on the SMILES string data set reaches 90.31%,which is 1.19%higher than the traditional GRU policy network.Using Cu₂（CO₂）₄ metal cluster and rhr topology,this system generates 40 MOFs,of which 26 MOFs have a greater hydrogen adsorption capacity than MOFs synthesized by chemical experiments with the same metal cluster and topology:2.776 mmol/g,of which the maximum hydrogen adsorption capacity is 6.181 mmol/g;using Zn₄O（CO₂）₆ metal cluster and pcu#1 topology,this system generates 40 MOFs,of which 31 MOFs have a greater hydrogen adsorption capacity than MOFs synthesized by chemical experiments with the same metal cluster and topology:1.714 mmol/g,of which the maximum hydrogen adsorption capacity is4.116 mmol/g.By analyzing the structural characteristics of MOFs with large hydrogen adsorption capacity obtained by reverse design,it can provide theoretical guidance for the chemical experiment method to synthesize MOFs with high hydrogen adsorption rate.