Chemical Industry and Engineering Progress ›› 2025, Vol. 44 ›› Issue (10): 5673-5688.DOI: 10.16085/j.issn.1000-6613.2024-1284

• Energy processes and technology • Previous Articles    

Analysis and forecasting of Chinese methanol price based on the intelligent chemical engineering large language model

WANG Wenyang1,2(), LUO Yuping1, YU Jiahuan1, ZHOU Jibin2, YE Mao2(), LIU Zhongmin2   

  1. 1.School of Maritime Economics and Management, Dalian Maritime University, Dalian 116026, Liaoning, China
    2.Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian 116023, Liaoning, China
  • Received:2024-08-05 Revised:2024-11-12 Online:2025-11-10 Published:2025-10-25
  • Contact: YE Mao

基于智能化工大模型的中国甲醇价格分析与预测

王文洋1,2(), 罗玉平1, 余佳洹1, 周吉彬2, 叶茂2(), 刘中民2   

  1. 1.大连海事大学航运经济与管理学院,辽宁 大连 116023
    2.中国科学院大连化学物理研究所,辽宁 大连 116023
  • 通讯作者: 叶茂
  • 作者简介:王文洋(1991—),副教授,硕士生导师,研究方向为统计分析、机器学习与人工智能等。E-mail:wangwenyang@dlmu.edu.cn
  • 基金资助:
    教育部人文社会科学基金(22YJC910011);中国博士后科学基金(2023M733444);辽宁省人工智能领域科技创新重点研发计划(2023JH26/10200012);国家自然科学基金(22308348);辽宁滨海实验室资助项目(LBLF-2023-01)

Abstract:

As a multi-purpose chemical product and low-carbon clean fuel, the price fluctuations of methanol impact the global chemical industry chain and energy market. However, existing time series forecasting methods fail to capture the non-stationary and high volatility characteristics of methanol prices. In order to accurately predict methanol price in China, this article originally proposes the CEGPT-Price Forecaster for Methanol (CEGPT-PF-M) model based on the first intelligent chemical engineering large language model in China. It first comprehensively integrates more than 2.9 million time series data in the public database from 27 fields related to the methanol market and transfer-trains the baseline CEGPT-PF-M; secondly, this paper applies the maximum mutual information coefficient algorithm to extract data from non-public commercial databases, 10900 index data that are highly related to Chinese methanol price are screened out, a private database is constructed, and the parameters of the CEGPT-PF-M model are fine-tuned based on this database to achieve the best prediction effect on Chinese methanol price; finally, in terms of factor analysis, this article builds an influencing factor index system based on a private database to analyze the impact of exogenous variables on Chinese methanol price from both macro and micro levels. Empirical results show that the accuracy, interpretability, and scalability of the CEGPT-PF-M model in the Chinese methanol price prediction task are significantly more reasonable than existing models. The research conclusions of this article provide a practical reference for methanol producers, coal suppliers, and policymakers, and also provide new perspectives and methods for chemical product price research.

Key words: methanol price forecasting, Transformer architecture, intelligent chemical engineering large language model, transfer learning, explainable AI

摘要:

甲醇作为一种多用途化工产品和低碳清洁燃料,其价格波动对全球化工产业链和能源市场具有重要影响,然而现有时间序列预测方法在捕捉甲醇价格的非平稳性和高波动性特征方面存在显著局限。为精准预测中国甲醇价格,本文基于国内首个智能化工大模型,首先全面整合公开数据库中与甲醇市场具有相关性的27个领域的290万余条时间序列数据,迁移训练首个用于甲醇价格预测的生成式预训练时间序列预测模型——生成式预训练甲醇价格预测(CEGPT-price forecaster for methanol,CEGPT-PF-M)模型;其次,本文应用最大互信息系数算法,从非公开商业数据库中筛选出10900条与中国甲醇价格高度相关的指标数据,构建私有数据库,并基于此数据库对CEGPT-PF-M模型进行参数微调,以实现对中国甲醇价格的最佳预测效果;最后,在影响因素分析方面,本文基于私有数据库构建影响因素指标体系,从宏观和微观双层面分析外生变量对中国甲醇价格的影响程度。结果表明,CEGPT-PF-M模型在中国甲醇价格预测任务中的准确性、解释性和可扩展性,均显著优于现有模型。本文的研究结论为甲醇生产商、煤炭供应商和政策制定者提供有效参考,同时也为化工产品价格研究提供新视角和新方法。

关键词: 甲醇价格预测, Transformer架构, 智能化工大模型, 迁移训练, 可解释AI

CLC Number: 

京ICP备12046843号-2;京公网安备 11010102001994号
Copyright © Chemical Industry and Engineering Progress, All Rights Reserved.
E-mail: hgjz@cip.com.cn
Powered by Beijing Magtech Co. Ltd