神经网络协同遗传算法高效筛选镍基甲烷化催化剂研究

Methanation catalysts and process are the key techniques for production of synthetic natural gas (SNG) from coal, during which, the combination of several metal promoters has been considered as an effective approach to improving the catalytic properties of Ni-based catalysts. However, due to it is time-consuming and lack of better understanding about the regularity in structure-activity relationships of a multi-component catalyst, the traditional screening with an experimental method often failed to accurate estimate and seek the optimal formulate from massive promoters. The fact makes it necessary to use intelligent methods for designing some highly efficient catalyst with desired properties. To achieve this goal, in this research, we propose a strategy for optimizing the experimental design of Ni-based catalysts modified by multiple promoters, which comprises the combination of an artificial neural network (ANN) and a genetic algorithm (GA), depending on the catalytic principle simultaneously, with the aim of increasing efficiency in catalyst screening and finally finding the optimization formulation of the resulting catalyst. The main components of this research proposal are as follows: The catalyst library with multiple promoters will be initially built to determine the optimized object followed by compiling appropriate GA; the ANN is to be adopted while the model for predicting the catalyst system is systematically established to realize multidimensional optimization in screening the efficient generation; the catalyst synthesis, characterization and activity evaluation of the final selected one will be experimentally conducted; clustering of the microstructure and reaction data is to be reasonably analyzed, the density function theory (DFT) will be applied to simulate the catalysis model; the reaction mechanism can be explained clearly and the quantitative relation between catalyst microstructure and its catalytic performance could be established theoretically; and the synergistic effect and the controlling factors of the promoted catalyst with dopants will be investigated in detail to make the variations of catalyst structure and improvement of catalytic properties available, which will lay scientific foundation to prepare highly efficient catalysts for methanation.

煤制天然气的核心技术为甲烷化催化剂及合成工艺，通过多种金属助剂的组合添加成为优化和改善镍基催化剂性能的重要途径。由于传统实验筛选方法复杂耗时，而且缺乏对多组分催化剂构效关系的规律性认识，想要从众多助剂中寻找最佳的配方是极其困难的。本项目根据组合催化技术原理，利用神经网络结合遗传算法优化多助剂镍基甲烷化催化剂，提高催化剂筛选效率，确定最佳催化剂的组成。主要研究内容包括：1）建立多助剂组分的初始催化剂库，确定优化目标，编写适合该体系的遗传算法；2）确定神经网络结构，建立催化剂预测模型，结合遗传算法实现多目标优选高效催化剂；3）进行催化剂实验制备、表征和活性评价，收集微观结构和反应活性数据；4）应用密度泛函理论模拟催化剂模型，阐明催化作用机理，建立微观结构与催化性能的定量关系；5）揭示多组分改善催化剂性能的协同效应及其调变规律，实现催化剂结构调控和性能优化，为制备高效的甲烷化催化剂奠定科学基础。

利用神经网络结合遗传算法优化镍基甲烷化催化剂，可以避开复杂费时的传统筛选过程，建立新型甲烷化催化剂筛选方法。优化筛选出能够提高镍基甲烷化催化剂活性、抗烧结性能和抗积碳性能的助剂及其组合用量，确定最佳的催化剂的组成，为制备高效的甲烷化催化剂提供技术参考。主要研究内容和结果如下：.（1）将数据挖掘技术用于调节镍基甲烷化催化剂活性助剂的筛选。通过主成分分析及K-均值方法对助剂及其活性数据处理后，建立了Ni/Al2O3甲烷化催化剂的组份模型，可以避开复杂的催化反应机理，预测元素Re是镍基甲烷化催化剂的最优助剂，实验验证了该预测的准确性；.（2）采用高斯过程回归建模甲烷化催化剂与催化剂抗烧结性之间的关系，并引入期望增量EI，进一步增加了该模型的准确性，筛选出了能有效改善抗烧结性能的助剂元素Cs。此外，使用虚拟元素法计算物化性质对催化性能的影响，并对催化剂的设计及开发具有一定的指导作用；.（3）将机器学习算法应用于提高Ni/Al2O3催化剂抗积碳性助剂的预测。利用主成分分析变换和高斯混合模型聚类对56种元素的16种理化性质进行了处理，然后选择12种代表性元素进行后续的抗积碳实验。根据实验结果和所选元素的主成分，分别建立了径向基函数网络、支持向量机和高斯过程回归模型。与其他模型相比，高斯过程回归模型的预测结果更加准确，成功筛选到最佳的抗积碳助剂元素W；.（4）将神经网络与改进的第二代遗传算法相结合（ANN-NSGA-II）用于建模优化Re，W，Cs助剂的最佳配比。通过构建多助剂组分与多性能的关系模型，对多目标组合添加的配方实现筛选优化，预测出的Ni-Re0.45W0.31Cs0.24/Al2O3催化剂样本，经实验验证，可以提高活性及改善抗烧结性和抗积碳性，其综合性能达到最佳，为新型高效催化剂的设计和筛选提供了一种科学有效的方法。