跳到正文
arXiv· Yiwei Fu·· 2 天前精选AI 评分58

两座燃气电厂部署本地 LLM 助手,把取证据交给固定软件

Deploying a Control-Logic LLM Assistant in Air-Gapped Power Plants: Lessons in Source Authority and Bounded Retrieval

AI 导读

作者报告在两座燃气电厂部署控制逻辑证据层的经验,该层支撑一套本地运行的操作员辅助系统。第一个原型用 VLM 从工程师日常交换的 PDF 打印件重建控制拓扑,在 6979 页文档上连接精度为 0.51,在 4xH200 GPU 上耗时 3.1 小时,随后发现编写系统本身有结构化导出。

变化判断

两座燃气电厂把控制逻辑问答交给本地模型,模型只负责听懂问题和解释结果,取证据交给固定软件,重复提问的 token 用量降到原来的二十分之一。

正文 · 原文

View PDF HTML (experimental)

Abstract:Deploying Large Language Model (LLM) assistants in industrial operational technology environments poses constraints largely absent from cloud applications: site policies restrict engineering and operational data from leaving the plant, on-premises compute is limited, and incorrect causal attribution can mislead incident response. We report lessons from deploying a control-logic evidence layer that underpins a broader on-premises operator assistance system at two gas power plants. Two prototypes revealed where generative models added avoidable uncertainty and cost. The first used a Vision Language Model (VLM) to reconstruct control topology from PDF printouts routinely exchanged by engineers. Across the full 6,979-page document, extraction reached 0.51 connection precision and required 3.1 hours on 4xH200 GPUs; consultation with controls engineers then revealed a native structured export from the authoring system. The second used an LLM tool-calling agent to plan repeatedly over eight predefined graph operations. In a controlled benchmark of 40 template-generated questions, replacing this agent with explicit execution plans reduced mean token use by 20x and median latency by 3x, with identical retrieved node sets across repeated runs. The deployed design ingests the authoritative export into a versioned knowledge graph, executes bounded graph operations in fixed software, and confines the local LLM to fallback intent classification and evidence-grounded explanation. At one site, the graph contains 221K nodes across seven controllers. Our experience suggests a practical division of labor for high-consequence AI: LLMs interpret natural-language requests and explain results, while deterministic software constructs and retrieves technical evidence grounded in authoritative sources. Abstention, entity disambiguation, graph versioning, and read-only integration provide additional safeguards.
Comments: Extended version of a paper accepted at IAAI-27 (AAAI 2027). 8 pages, 2 figures, 3 tables
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.20398 [cs.AI]
  (or arXiv:2611.20398v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.20398

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Yiwei Fu [view email]
[v1] Mon, 16 Nov 2026 01:54:11 UTC (32 KB)

来源:arXiv · arxiv.org