石油化工高等学校学报 ›› 2025, Vol. 38 ›› Issue (6): 40-48.DOI: 10.12422/j.issn.1006-396X.2025.06.005

• 石油工程 • 上一篇    下一篇

融合外生变量的储气库短期供气量预测研究

刘倩1,2(), 侯磊1,2(), 喻鹏飞1,2, 王敏聪1,2, 李睿麒3   

  1. 1.中国石油大学(北京) 机械与储运工程学院,北京 102249
    2.国家能源地下储气库研发中心,北京 102249
    3.国家石油;天然气管网 浙江省天然气管网有限公司,浙江 丽水 323000
  • 收稿日期:2025-06-10 修回日期:2025-07-07 出版日期:2025-12-25 发布日期:2025-12-07
  • 通讯作者: 侯磊
  • 作者简介:刘倩(2001⁃),女,硕士研究生,从事储气库群协同调峰及运行优化方面的研究;E⁃mail:lq010201@163.com
  • 基金资助:
    国家自然科学基金面上项目(52174063)

Study on Short⁃Term Gas Supply Prediction of Gas Storage Incorporating Exogenous Variables

Qian LIU1,2(), Lei HOU1,2(), Pengfei YU1,2, Mincong WANG1,2, Ruiqi LI3   

  1. 1.College of Mechanical and Transportation Engineering,China University of Petroleum (Beijing),Beijing 102249,China
    2.National Energy Research and Development Center of Underground Gas Storage,Beijing 102249,China
    3.Zhejiang Natural Gas Pipeline Co. ,Ltd. ,China Oil & Gas Pipeline Network Corporation,Lishui Zhejiang 323000,China
  • Received:2025-06-10 Revised:2025-07-07 Published:2025-12-25 Online:2025-12-07
  • Contact: Lei HOU

摘要:

通过预测储气库短期供气量,并将其作为储气库季节性调峰量,可在保障供气可靠性的同时,提升储气设施运行效率与经济效益,破解由季节性“峰谷差”导致的供需失衡难题。通过精准预测下游用户的天然气需求量,能够有效反映储气库在短期内所需的供气量。选取某地区2021-2024年的每日天然气用气量,结合气温变化及日期类型,综合考量趋势项、季节项及节假日效应,提出了适用于储气库短期供气量预测的Prophet预测模型;采用平均绝对误差(rMAE)、平均绝对百分比误差(rMAP)、均方根误差(rRMS)、决定系数(R2)4项性能指标,对比评估了Prophet模型与STL分解、VARMAX等5种常用预测模型的预测性能。结果表明,Prophet模型在测试集上的rMAE为13.15 m3rMAP为2.71%,rRMS为16.52 m3R2为0.99,显著优于其他模型;在冬季用气高峰期,其预测误差可控制在5%以内。Prophet模型通过融合气候条件、日期类型两个外生变量,能够准确捕捉下游用户天然气用气量的季节性与突发性波动特征,提高储气库供气量预测精度,为储气库调峰保供提供关键数据支持。

关键词: 供气量预测, Prophet模型, 时间序列模型, 深度学习, 机器学习

Abstract:

This study uses short?term gas supply prediction for storage facilities as the seasonal peak?shaving volume.This approach ensured supply reliability while improving the operational efficiency and economic benefits of storage facilities,tackling the supply?demand imbalance caused by seasonal peak?valley differences.Accurate prediction of downstream users' natural gas demand could effectively reflect the required short?term gas supply from storage.Daily natural gas consumption data from a specific region during 2021-2024 was selected.Incorporating temperature variations and date types,the study comprehensively considers trend components,seasonal patterns, and holiday effects.A Prophet forecasting model suitable for predicting short?term gas supply from storage was proposed.Four performance metrics?Mean Absolute Error (rMAE),Mean Absolute Percentage Error (rMAP),Root Mean Square Error (rRMS),and Coefficient of Determination (R2)?were used to comparatively evaluate the Prophet model against five other common models (including STL decomposition and VARMAX).The results show that the Prophet model achieves an rMAE of 13.15 m3,an rMAP of 2.71%,an rRMS of 16.52 m3,and an R2 of 0.99 on the test set,which is significantly superior to other models.During the winter gas consumption peak period,its prediction error can be controlled within 5%.By integrating two exogenous variables?climatic conditions and date types.The Prophet model can accurately capture the seasonal and sudden fluctuation characteristics of natural gas consumption of downstream users, improve the prediction accuracy of gas storage supply,and provide key data support for peak?shaving and supply guarantee of gas storage reservoirs.

Key words: Gas supply forecast, Prophet model, Time ? series model, Deep learning, Machine learning

中图分类号: 

引用本文

刘倩, 侯磊, 喻鹏飞, 王敏聪, 李睿麒. 融合外生变量的储气库短期供气量预测研究[J]. 石油化工高等学校学报, 2025, 38(6): 40-48.

Qian LIU, Lei HOU, Pengfei YU, Mincong WANG, Ruiqi LI. Study on Short⁃Term Gas Supply Prediction of Gas Storage Incorporating Exogenous Variables[J]. Journal of Petrochemical Universities, 2025, 38(6): 40-48.

使用本文