基于机器学习的铜基二元纳米合金表面CO2电催化机理研究

The research of realizing the carbon recycling utilization with the strategy of CO2 reduction catalysis has great scientific significance and application value. Now it is still not very crystal clear for CO2 electrocatalytic active site and reaction mechanism on nano-scale bimetallic alloys surface. Another problem is that the established models of electrode surface in theoretical simulation are always too simple to describe the out-of-flatness surface in realistic electrode on atomic level. For solving these two problems, my project will choose the Cu-based nano-scale bimetallic alloys, one of most promising alloy series, employing machine-learning (ML) scheme to research CO2 electrocatalytic activity mechanism and catalytic enhancement method on its large scale complicated surface. In comprehensive consideration of the effects of atomic steps, edges, corners and defects on catalytic alloy surface, first-principles calculation and smooth overlap atomic position (SOAP) method are utilized to construct the effective and adaptive training-set. Then Bayesian linear regression method is applied to find the relationship between structural geometries and energy from the prepared training-set with ML big data calculation. And this relationship is described as a parameter set, which can calculate the adsorption energy of any surface position in the large-scale complicated surface model. Next, the adsorption energy can be calculated as activation energy of transitional status in catalysis simulation process by the Brønsted–Evans–Polanyi (BEP) formula, and the activation energy will be transferred as catalytic turnover frequency (TOF) by Sabatier formula. Thus, these two formulae construct the relationship between adsorption energy and catalytic velocity. In summary, this project will systematically research the way that element types, surface geometries and element ratio affect the distribution and intensity of the catalytic active site, revealing the electrode surface characteristics of high electrocatalytic efficiency to CO2 reduction. The studies of this project are expected to provide theoretical guidance for developing novel CO2 catalytic materials in industrial application, giving a feasible strategy to design and realize that goal.

通过CO2还原催化实现碳元素的循环使用，具有十分重要的科学意义与应用价值。目前二元纳米合金表面CO2电催化活性位点和反应机制还不是十分清楚，常用的电极表面模型过于简单。本项目选用具有应用前景的铜基二元纳米合金，采用机器学习方法研究其大尺度复杂表面的CO2电催化活性机制与催化增强方法。在综合考虑原子台阶、边界、拐角、缺陷等对催化表面影响的基础上，通过第一性原理计算和光滑重叠原子位置法构建高效泛化训练集。再使用贝叶斯回归进行机器学习大数据处理，拟合出表面上的“结构-能量”关联参数集,该参数集可计算大尺度材料表面的吸附能。最后由BEP公式和Sabatier公式发展出“吸附能-催化速率”关联模型。本项目将系统地研究元素、结构、组分是如何影响催化活性位点的分布和强度，揭示高效催化电极的表面特征。期望对开发可工业化应用的新型高效CO2催化材料提供理论指导并给出可行的实现途径。

通过CO2还原催化实现碳元素的循环使用，具有十分重要的科学意义与应用价值。但人们对二元纳米合金表面的CO2电催化反应机制仍不能完全确定，这也限制了高效催化材料的研制开发。本项目选用具有应用前景的铜基二元纳米合金，采用机器学习方法研究其大尺度复杂表面的CO2电催化活性机制与催化增强方法。在综合考虑原子台阶、边界、拐角对催化表面影响的基础上，通过第一性原理计算和光滑重叠原子位置法(SOAP)构建高效泛化训练集。. 按照项目计划，我们实现了以机器学习程序预测合金形成能、CO等小分子的吸附能、合金催化速率。根据不同的合金类型、组分比例、晶面、原子排布，我们总共排列组合得到了400余个二元合金表面结构。将每个表面结构的能量、SOAP函数值、以及相关的描述符作为训练集参数代入机器学习模型。模型包括线性回归、随机梯度、多项式、Ridge回归、Lasso回归等方法，我们分别测试并获得了对应的参数集。为了研究CO2还原催化的关键决速步骤CO加氢成为CHO，我们研究了CO在合金表面模型的吸附能。经过测试集的初步验证，人工神经网络被发现为较优的机器学习算法。之后根据Brønsted–Evans–Polanyi（BEP）公式和Sabatier公式，我们建立了吸附能与催化速率的关系，完成了二元合金表面模型上催化转换频率的计算。基于本项目的理论计算指导，我们制备了单原子Pt-NiO/Ni催化剂，在碱性溶液中表现出了杰出的催化电解水产氢性能，比商业Pt/C催化剂高41倍。. 总的来说，本项目的工作尝试解决了第一性原理计算只能处理小体系的问题。本研究将第一性原理计算与机器学习结合，实现模拟了大体系、多组分、多种表面形态的二元合金电极形成能和CO吸附能。创新地发展机器学习模型研究大尺度纳米合金表面电催化的方法。本研究可开发为新型纳米合金电催化材料提供了理论指导，提高研发效率。

内容获取失败，请点击重试