模拟器接地大语言模型用于工业因果推理：污水处理决策支持的工具使用、结构化注入与工厂可移植检索
原标题：Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support
AI 摘要
该研究比较了三种将冻结的Qwen2.5-32B-Instruct模型与污水处理模拟器（CCSS-IX）结合的方法：实时模拟器工具调用、结构化参数注入和DRR检索器。在198个因果问题上，三种方法分别达到99.5%、79%和75.8%的准确率，均优于基线RAG的48%。DRR检索器仅110M参数，训练时间约17秒，且能跨工厂迁移，在反事实基准上表现最佳。该机制在ARC基准上达到79%，优于Llama-3.1-8B的76%，表明其通用性。
正文节选
Computer Science > Computation and Language Title:Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support View PDF HTML (experimental) Abstract:Wastewater operators need answers grounded in how their plant's variables interact and how fast effects propagate, not in generic pretraining text, when asking causal questions such as "why is N2O rising?" or "what happens