用模拟器增强大模型，让污水处理决策更精准可靠。
Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support
- 用模拟器实时响应、参数注入和可迁移检索三种方式增强模型因果推理能力。
- 最佳方法在因果问答上准确率达99.5%，在反事实推理上比基线高16.3个百分点。
- 模型可在不同工厂快速迁移，适合工业界需要可解释、可部署的智能决策系统。
污水处理操作员在面对'为何一氧化二氮上升？'或'若减少曝气20%会怎样？'等因果问题时，需要基于其厂区变量交互与影响传播速度的答案，而非通用预训练文本。本文对比三种将冻结的Qwen2.5-32B-Instruct模型与可解释的污水厂模拟器（CCSS-IX）结合的方法：实时模拟器调用（方法1）、结构化参数注入（方法2）和解耦检索-推理（DRR）检索器（方法3）。在198个因果问题基准上，三者准确率分别为99.5%、79%和75.8%，高于最强检索增强基线（48%）。DRR检索器仅110M参数，每厂训练约17秒；跨厂迁移至生物特性不同的厂区仍达88%，而方法2的静态表无法迁移。在60个反事实问题上，仅方法3能回答干预后变化，较方法2提升16.3个百分点（95%置信区间[+7.1, +26.4]），且在时间尺度与运行工况类别上达到100%准确。在包含开放书本知识的AI2推理挑战（ARC）上，相同检索机制达79%，优于未约束的Llama-3.1-8B（76%）和全注入方法（74%），验证其泛化能力。本文首次在单一模拟器下系统比较了实时工具使用、静态参数注入与学习型数值检索在工业因果推理中的表现。
原文摘要 · Abstract (English)
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 if I cut aeration by 20%?". We compare three concrete ways to ground a frozen Qwen2.5-32B-Instruct model in an architecturally interpretable wastewater simulator (CCSS-IX): a live simulator oracle (Method 1), structured parameter injection (Method 2), and a Decoupled Recall-Reasoning (DRR) retriever (Method 3). On a 198-question causal benchmark the three reach 99.5%, 79%, and 75.8%, forming a deployment ladder above the strongest retrieval-augmented baseline at 48%. The DRR retriever has 110M parameters and trains per plant in ~17 seconds; after cross-plant transfer to a biologically distinct plant it still reaches 88%, while Method 2's static table cannot transfer. On a 60-question counterfactual benchmark only Method 3 handles queries about what happens after an intervention: +16.3 pp over Method 2, paired 95% CI [+7.1, +26.4] pp, with 100% on the timescale and operating-regime categories. On the AI2 Reasoning Challenge (ARC) with an OpenBookQA fact corpus, the same selective-retrieval mechanism reaches 79% versus unconstrained Llama-3.1-8B 76% and full-injection 74%, a +3 pp out-of-domain replication that argues against a result specific to wastewater treatment. We provide the first single-simulator comparison of live tool-use, static parameter injection, and learned numerical-parameter retrieval for industrial causal question answering.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。