关键词: 石油化工, 高质量数据集, 大语言模型, 人工智能, 过程系统, 模型, 化学反应器

Abstract:

The rapid advancement of generalized large language models (LLMs) and the maturation of the open-source ecosystem are driving artificial intelligence (AI) toward deeper integration with industrial domains. As a typical process industry, the petrochemical industry has accumulated vast amounts of raw data over long-term operations. These data are often fragmented, unstructured, and poorly labeled, making them difficult to use directly for training LLMs. Such limitations constrain LLMs' ability to understand and apply specialized petrochemical knowledge, leading to frequent hallucinations that hinder their deployment. Focused on the key to applying LLMs in the petrochemical industry — high-quality industry datasets, a systematic review is presented on the challenges and opportunities in constructing such datasets. Based on the characteristics of industry data, a general framework is proposed for building and applying high-quality datasets in the petrochemical industry, aiming to establish a scalable and practical dataset system. The prospects and development directions of dataset-driven industrial applications are also discussed to promote the deep integration of LLMs with the petrochemical industry.

Key words: petrochemical industry, high-quality datasets, large language model, artificial intelligence, process systems, model, chemical reactors