跳转至

文章背景与核心概要

传统的工业级Actor-Critic（演员-评论家）强化学习方法通常将连续动作视为匿名的数值坐标，迫使系统通过反复试错来学习特定动 ACTION 如何影响过程变量、影响方向以及延迟时间。然而，现有的工业固定文档中其实已经明确记录了这些物理和工艺关系，但这些信息通常以开放文本的形式存在，无法直接映射到数值化的策略或当前的实际运行工况中。

为了解决这一痛点，本文提出了 LCAE（LLM-Guided Contextual Action Evaluation，大模型引导的上下文动作评估） 框架。该方法在训练前利用大语言模型（LLM）对标准文档进行解析和规范化，将其转化为冻结的“动作-观测-方向-延迟”关系基底。在运行过程中，最近的数值型“动作-响应”历史会动态调制这些关系强度，从而使评论家能够通过状态条件化的非关联系数来评估动作。最重要的是，LLM 或嵌入模型在训练或部署期间均不在线运行，这确保了最终的策略可以完全依赖冻结的语义工件和可见的数值历史。

LLM-Guided Contextual Action Evaluation for Operational Decisions in Industrial Processes

Summary

Industrial actor-critic methods traditionally treat continuous actions as anonymous numerical coordinates, forcing the system to learn through trial and error how specific actions influence process variables. While fixed industrial documentation already outlines these relationships, this information exists as open-text statements that do not directly map to numerical policies or current operating conditions.

This paper introduces LCAE (LLM-Guided Contextual Action Evaluation), a novel framework that uses a Large Language Model (LLM) prior to training to parse and normalize standard documentation into a frozen action–observation–direction–delay relation basis. During operation, recent numerical action–response history modulates the strength of these relations, enabling a critic to evaluate actions through a state-conditioned, nonlinear action-effect field. Crucially, neither the LLM nor the embedding model runs online during training or deployment—allowing the final policy to rely exclusively on frozen semantic artifacts and visible numerical history.

Metadata & Reference Information

Metadata & Reference Information

  • arXiv ID: arXiv:2608.24156
  • Primary Subject: Systems and Control (eess.SY)
  • Secondary Subjects: Artificial Intelligence (cs.AI), Emerging Technologies (cs.ET)
  • Submission Date: August 25, 2026
  • DOI: 10.48550/arXiv.2608.24156

Authors

Authors

  • Youcheng Zong
  • Runda Jia
  • Dakuo He

Abstract

工业级 Actor-Critic 方法通常将连续动作表示为匿名的数值坐标。因此，它们必须从有限的交互中学习每个动作影响哪些过程变量、影响方向如何以及延迟多久。现有的工业文档已经描述了部分此类关系，但其开放文本表述既不反映当前的运行工况，也无法直接契合数值策略。

Industrial actor--critic methods usually represent continuous actions as anonymous numerical coordinates. They must therefore learn from limited interactions which process variables each action affects, in which direction, and after what delay. Fixed industrial documents already describe part of these relations, but their open-text statements neither represent the current operating condition nor directly fit a numerical policy.

本文提出了工业过程操作决策的大模型引导上下文动作评估方法（LLM-Guided Contextual Action Evaluation for Operational Decisions in Industrial Processes, LCAE）。该方法在训练前利用大语言模型将固定文档规范化为冻结的“动作-观测-方向-延迟”关系基底。随后，近期的数值型“动作-响应”历史会调制每种关系的当前强度，而被评估的动作则在同一基底中形成一个状态条件化的非线性动作效应场。评论家通过该场评估动作，而演员则利用相同的关系增益来生成动作，从而将文档语义融入最大熵策略学习中。在训练或部署期间，LLM 或嵌入模型均不在在线运行；部署的策略仅使用冻结的语义工件和可见的数值历史。

This article presents LLM-Guided Contextual Action Evaluation for Operational Decisions in Industrial Processes (LCAE), which uses a large language model before training to normalize fixed documents into a frozen action--observation--direction--delay relation basis. Recent numerical action--response history then modulates the current strength of each relation, while the evaluated action forms a state-conditioned nonlinear action-effect field in the same basis. The critic evaluates actions through this field, and the actor uses the same relation gains to generate actions, making document semantics part of maximum-entropy policy learning. Neither the LLM nor the embedding model runs online during training or deployment; the deployed policy uses only frozen semantic artifacts and visible numerical history.

该方法提出了一个可证伪的假设：当记录的关系正确且近期历史反映了其上下文强度时，这种动作表示应能提供比原始动作坐标更有价值的决策偏差。

The method states a falsifiable hypothesis: when documented relations are correct and recent history reflects their contextual strength, this action representation should provide a more useful decision bias than raw action coordinates.