arXiv · cs.AI · Guoyang Shi, Siqi Ding, Xuanhe Wang, Pei Guo, Changquan Yu, Jiale Hu, Jiayi Zhi, Zhengyuan Chen, Jianguo Chen, Tianyuan Liu · 5 小时前模型评分35
从预测到反馈：评估学习动力学在托卡马克控制中的作用
From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control
导读
研究用校准的自由边界 Grad-Shafranov 求解器收集四组数据集，训练16个动作条件 Transformer，并在5k替代模型上优化PPO策略后直接评估闭环表现。结果显示仅联合覆盖的D4跨条件平均误差单调下降，D1分布内改善但分布外退化；动作噪声训练的策略仅完成5/24条件，而标称训练完成19/24。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
应用领域：能源、科学研究
事件类型：研究发布
证据摘录
- “Learned temporal models are moving beyond offline forecasting toward interactive environments for decision making, yet logged-trajectory accuracy need not predict closed-loop behavior under policies optimized in the learned model.”
- “Based on one trained policy per configuration and one evaluation realization per condition, the results support combining prediction error, episode completion, condition-level duration, and tracking quality.”
来源：arXiv · cs.AI · arxiv.org