SteelAgent：LLM 编排的物理信息钢材性能预测与泛化审计
原标题：SteelAgent: LLM-Orchestrated, Physics-Informed Steel Property Prediction with Audited Generalization and Calibrated Uncertainty
AI 摘要
该论文提出 SteelAgent，一个由大语言模型编排、结合物理冶金描述符与机器学习模型的钢材性能预测系统，支持性能预测、不确定性报告和逆向设计。研究重点考察不同留出单元对评估结论的影响：梯度提升模型在随机划分下 R² 为 0.859，但按钢种家族留出时降至 0.113；名义 90% 预测区间的实际覆盖率也从 91.0% 降至 37.6%。物理特征的效果随评估人群而变，说明报告性能估计时应同时说明评估人群。
正文节选
SteelAgent: LLM-Orchestrated, Physics-Informed Steel Property Prediction with Audited Generalization and Calibrated Uncertainty Note: A shorter version of this work, SteelAgent: An LLM-Orchestrated System for Physics-Informed Steel Property Prediction and Generalization Auditing, was published in the Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence (IJCAI-26), Demonstrations Track, pp. 8547–8550 (2026). doi:10.24963/ijcai.2026/1000. Abstract Steel propert