当前位置：首页 > 过刊浏览->2026年第52卷第3期
基于大语言模型的人工智能HSE履职能力评估系统开发与应用
Development and application of artificial intelligence HSE performance capability evaluation system based on large language model
浏览(15) 下载(0)
- DOI:
- 作者:
- 徐亚洲，胡云鹏，尚芳兰，刘永奇，高帅
- 作者单位:
- 国家管网集团甘肃公司，甘肃兰州 730070
- 关键词:
- HSE管理体系；HSE履职能力；AI智能体；大语言模型
HSE management system; safety HSE responsibility competency; AI agent; large language model
- 摘要:
- 为解决管网企业健康、安全、环境管理中HSE履职能力评估环节存在的人工组织成本高、评估标准不
统一、评估效率低的难题，依托大语言模型，在国家管网集团WeAgent平台（一站式AI应用开发和部署平台）
上设计并开发了一套人工智能HSE履职能力评估系统。系统的核心实现路径分为四步：首先，运用自然语言处
理技术与向量检索技术，构建结构化、全覆盖的安全环保领域知识库，为评估提供标准化知识支撑；其次，以
大语言模型作为推理引擎，通过设计结构化提示词规范大语言模型在各个评估环节的行为逻辑，实现对评估信
息的精准理解、逻辑推理与评估内容生成；再次，通过多层级业务逻辑工作流编排，将复杂的评估任务分解为
任务初始化、动态提问、智能追问、结果分析、报告生成等一系列自动化子任务，保障评估流程的有序推进；
最后，借助智能体技术，将上述能力封装为可自主运行的智能交互界面，自动生成包含履职能力短板分析及个
性化改进建议的完整评估报告。研究的主要创新在于，建立了一套融合领域知识库、结构化工作流与AI智能体
的HSE履职能力自动化评估框架，并通过工程实践验证了该系统替代传统人工评估模式的可行性。这一成果不
仅解决了管网企业HSE评估环节的传统难题，更为油气及能源管网类企业的HSE管理智能化提供了可复制、
可推广的实践路径。
To address the engineering management challenges in the health, safety, and environment management of pipeline enterprises, such as high manual organization costs, inconsistent assessment standards, and low assessment efficiency in the assessment link of safety and environmental responsibility competency, an artificial intelligence assessment system was designed and developed based on the large language model on the WeAgent platform of PipeChina. The core implementation path of the system was divided into four steps: First, by applying natural language processing and vector retrieval technologies, a structured and comprehensive safety domain knowledge base was constructed to provide standardized knowledge support for the assessment. Second, by taking the large language model as the inference engine, structured prompts were designed to regulate the behavioral logic of the large language model in each assessment link, realizing the accurate understanding, logical reasoning, and information generation of assessment information. Meanwhile, through multi-level business logic workflow orchestration, complex assessment tasks were divided into a series of automated subtasks, such as initialization, dynamic questioning, intelligent follow-up questioning, result analysis, and report generation, ensuring the orderly advancement of the assessment process. Finally, with the aid of agent technology, the above capabilities were encapsulated into an independently operable interactive interface, which automatically generates an assessment report containing capability gap analysis and personalized improvement suggestions. The main innovation of this study lies in proposing an automated HSE responsibility competency assessment framework integrating a domain knowledge base, structured workflows, and AI agents, and engineering practice verifies the feasibility of this system replacing traditional manual assessments. This achievement not only solves traditional management problems in the HSE assessment link of pipeline enterprises but also provides a replicable and promotable practical path for intelligent safety management in similar enterprises.