正文：
近日，来自厦门大学、深圳大学、武汉大学、南京航空航天大学和英国利物浦大学的研究团队在材料科学领域取得重要突破，他们开发了一种基于迁移学习的范式，成功预测并合成了高性能钙钛矿氧化物电催化剂。这项研究成果已于7月26日发表在《Nature Communications》上，为加速发现和开发用于析氧反应（OER）的高性能电催化剂提供了新思路。
传统材料发现依赖于反复试验或偶然发现，效率低下且成本高昂。AI在发现新型催化剂方面的潜力巨大，但受到算法选择和数据质量与数量的影响。该团队开发了一种迁移学习范式，结合了预训练模型、集成学习和主动学习，能够预测未被发现的钙钛矿氧化物，并增强该反应的通用性。
研究人员通过筛选16,050种成分，鉴定和合成了36种新的钙钛矿氧化物，其中包括13种纯钙钛矿结构。其中，Pr0.1Sr0.9Co0.5Fe0.5O3（PSCF）和Pr0.1Sr0.9Co0.5Fe0.3Mn0.2O3（PSCFM）在10 mA cm^-2时分别表现出327 mV和315 mV的低过电位。电化学测量表明，这两种材料中O-O耦合的吸附质演化机制（AEM）和晶格氧机制（LOM）共存。
该研究团队提出的迁移学习范式包括七个步骤：数据提取、阳离子编码、特征嵌入、聚类、局部预测、全局集成和主动学习闭环实验验证。由于OER钙钛矿氧化物数据有限，研究人员还收集了非OER钙钛矿氧化物的数据，将数据集从94个条目扩展到140个条目，增加了48.9%。这种方法促进了不同材料系统之间的知识转移，从而显著提高了预测准确性。
实验验证和主动学习进一步验证了预测结果。研究人员从超过500万个预测点中选择了30种化学式进行实验验证，并成功合成了高性能的钙钛矿氧化物电催化剂。这一研究成果不仅为材料科学领域提供了新的研究工具和方法，也为实现绿色化学品的生产、推动碳中和进程提供了强有力的技术支持。
英语如下：
News Title: “Xiamen University Team’s Discovery in a Nature Journal: AI Accelerates the Development of High-Performance Perovskite Oxide Electrocatalysts”
Keywords: Xiamen University team, Nature journal, Perovskite oxide
News Content:
Recently, a research team from Xiamen University, Shenzhen University, Wuhan University, Nanjing University of Aeronautics and Astronautics, and the University of Liverpool, UK, made an important breakthrough in the field of materials science. They developed a paradigm based on transfer learning that successfully predicted and synthesized high-performance perovskite oxide electrocatalysts. This research outcome was published on July 26 in Nature Communications, offering a new approach for accelerating the discovery and development of high-performance electrocatalysts for the oxygen evolution reaction (OER).
Traditional material discovery relies on repeated trials or accidental discoveries, which are inefficient and costly. The potential of AI in discovering new catalysts is enormous, but it is limited by the choice of algorithms and the quality and quantity of data. The team developed a transfer learning paradigm that combines pre-trained models, ensemble learning, and active learning, capable of predicting undiscovered perovskite oxides and enhancing the generality of the reaction.
Researchers screened 16,050 compositions and identified and synthesized 36 new perovskite oxides, including 13 pure perovskite structures. Among them, Pr0.1Sr0.9Co0.5Fe0.5O3 (PSCF) and Pr0.1Sr0.9Co0.5Fe0.3Mn0.2O3 (PSCFM) exhibited low overpotentials of 327 mV and 315 mV, respectively, at 10 mA cm^-2. Electrochemical measurements showed that both materials featured coexisting O-O coupled adsorption evolution mechanisms (AEM) and lattice oxygen mechanisms (LOM).
The research team’s proposed transfer learning paradigm includes seven steps: data extraction, cation encoding, feature embedding, clustering, local prediction, global integration, and active learning closed-loop experimental verification. Due to the limited data of OER perovskite oxides, researchers also collected data from non-OER perovskite oxides, expanding the dataset from 94 entries to 140 entries, an increase of 48.9%. This method promoted knowledge transfer between different material systems, significantly improving prediction accuracy.
Experimental verification and active learning further validated the predictive results. Researchers selected 30 chemical formulas from over 5 million predicted points for experimental verification and successfully synthesized high-performance perovskite oxide electrocatalysts. This research not only provides new research tools and methods for the materials science field but also offers powerful technical support for the production of green chemicals and the realization of carbon neutrality.
【来源】https://www.jiqizhixin.com/articles/2024-07-31-3
Views: 2