Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support

Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support

基于模拟器的大型语言模型在工业因果推理中的应用:废水处理决策支持中的工具使用、结构化注入与工厂可移植检索


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 if I cut aeration by 20%?”.

摘要: 当废水处理操作员提出诸如“为什么 N2O 浓度在上升?”或“如果我将曝气量减少 20% 会发生什么?”这类因果问题时,他们需要的答案必须基于其工厂变量之间的相互作用方式以及影响传播的速度,而非通用的预训练文本。

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).

我们比较了将冻结的 Qwen2.5-32B-Instruct 模型植入架构可解释的废水模拟器(CCSS-IX)中的三种具体方法:实时模拟器预言机(方法 1)、结构化参数注入(方法 2)以及解耦检索-推理(DRR)检索器(方法 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%.

在一个包含 198 个问题的因果基准测试中,这三种方法的准确率分别达到 99.5%、79% 和 75.8%,均优于最强的检索增强基线(48%),形成了一个部署阶梯。

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.

DRR 检索器拥有 1.1 亿个参数,针对每个工厂的训练时间约为 17 秒;在跨工厂迁移至生物特性不同的工厂后,其准确率仍能达到 88%,而方法 2 的静态表则无法实现这种迁移。

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.

在一个包含 60 个问题的反事实基准测试中,只有方法 3 能够处理关于干预后结果的查询:相比方法 2 提升了 16.3 个百分点(配对 95% 置信区间 [+7.1, +26.4]),在时间尺度和运行工况类别上达到了 100% 的准确率。

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.

在结合 OpenBookQA 事实语料库的 AI2 推理挑战(ARC)中,同样的选定检索机制达到了 79% 的准确率,而无约束的 Llama-3.1-8B 为 76%,全注入法为 74%。这一跨领域复制实验提升了 3 个百分点,证明了该结果并非仅适用于废水处理领域。

We provide the first single-simulator comparison of live tool-use, static parameter injection, and learned numerical-parameter retrieval for industrial causal question answering.

我们首次在单一模拟器环境下,对用于工业因果问答的实时工具使用、静态参数注入和学习型数值参数检索进行了对比研究。