Machine Learning for Design and Full Chain Research of High‐Entropy Na‐Ion Cathodes

How to proficiently and accurately explore the vast compositional space of materials and accelerate the development of new materials with outstanding properties, especially structurally complex high-entropy oxides (HEOs), remains a challenge in materials science. To address this, a state-of-the-art hybrid flow machine learning (HFML) framework is proposed, which combines ensemble learning, unsupervised learning, and Bayesian optimization, enabling efficient discovery of hidden patterns and comprehensive exploration of the component space. Based on the proposed HFML, a new HEO cathode material is screened out from over 2 million candidates for sodium-ion batteries (Na0.95Li0.06Ni0.25Cu0.05Fe0.1Co0.05Mn0.44Ti0.05O2), which shows excellent cycling stability (capacity retention of 83.6% after 1200 cycles) and high-rate performance (110 mAh g-1 at 10 C, and 96 mAh g-1 at 20 C). Key factors affecting structural stability are identified, including s-block metal ions, Cu2+, high-valence d0 and d10 metal ions, and are verified by electrochemical tests and in situ X-ray diffraction (XRD) measurements. Additionally, pilot-scale production is achieved, and 2 Ah pouch cells based on this cathode demonstrate 95.0% capacity retention after 600 cycles. This work accomplishes the full chain study from artificial intelligence material prediction, creation, and performance verification to pilot application (mass production and pouch sodium batteries).