发电技术 ›› 2025, Vol. 46 ›› Issue (1): 9-18.DOI: 10.12096/j.2096-4528.pgt.24047

• 综合能源 • 上一篇    下一篇

基于组合相关系数的综合能源系统多元负荷短期预测研究

涂伟权, 李辉, 万铮   

  1. 上海电力大学自动化工程学院,上海市 杨浦区 200090
  • 收稿日期:2024-03-25 修回日期:2024-06-30 出版日期:2025-02-28 发布日期:2025-02-27
  • 通讯作者: 李辉
  • 作者简介:涂伟权(1998),男,硕士研究生,研究方向为新能源及电力负荷预测,twq5897@outlook.com
    李辉(1979),男,博士,副教授,研究方向为新能源微电网控制,本文通信作者,lihui@shiep.edu.cn
    万铮(1999),男,硕士研究生,主要研究方向为电力负荷的预测与调度,1229964672@qq.com
  • 基金资助:
    上海市科委重点项目(20dz120610)

Research on Short-Term Forecasting of Multiple Loads in Integrated Energy Systems Based on Composite Correlation Coefficients

Weiquan TU, Hui LI, Zheng WAN   

  1. College of Automation Engineering, Shanghai University of Electric Power, Yangpu District, Shanghai 200090, China
  • Received:2024-03-25 Revised:2024-06-30 Published:2025-02-28 Online:2025-02-27
  • Contact: Hui LI
  • Supported by:
    Key Projects of Shanghai Science and Technology Commission(20dz120610)

摘要:

目的 保证电网稳定运行,需准确预测综合能源系统中多种负荷。为此,针对现行算法预测精度低和多元负荷之间耦合性等问题,提出一种基于组合相关系数(composite correlation coefficient,CCC)的预测算法。 方法 首先,采用Savitzky-Golay(SG)滤波器和麻雀搜索算法(sparrow search algorithm,SSA)优化变分模态分解(variational mode decomposition,VMD),处理原始数据的特征;然后,利用组合相关系数分析耦合性,通过人工提取关联性大的一些特征输入结合编解码器(encoder-decoder,ED)的门控循环单元(gated recurrent unit,GRU)神经网络中,得到多元负荷预测结果;最后,基于前述算法,通过算例进行实验验证。 结果 相比于传统的GRU、长短期记忆网络、双向门控循环单元预测模型,所提算法的电负荷的预测误差分别降低了0.55%、0.49%、0.17%,冷负荷预测误差分别降低了0.71%、0.54%、0.19%,热负荷的预测误差分别降低了0.62%、0.55%、0.19%。 结论 所提算法模型能更好地捕捉关键信息,相比于其他算法有一定的优势,提高了预测的准确性。

关键词: 综合能源系统, 多元负荷预测, 能源耦合, 组合相关系数, 编码器-解码器, 门控循环单元, 滤波

Abstract:

Objectives To ensure the stable operation of the power grid, it is crucial to accurately predict the multivariate loads within an integrated energy system. To address the issues of low prediction accuracy and the coupling between multivariate loads in current algorithms, a predictive algorithm based on the composite correlation coefficient (CCC) is proposed. Methods The Savitzky-Golay (SG) filter and sparrow search algorithm (SSA) are used to optimize variational mode decomposition (VMD) for processing the characteristics of raw data. Subsequently, the CCC is utilized to analyze coupling, and highly correlated features are manually extracted and input into a gated recurrent unit (GRU) neural network combined with an encoder-decoder (ED) structure to obtain multivariate load prediction results. Finally, based on the aforementioned algorithm, experimental verification is carried out using numerical examples. Results Compared with traditional GRU, long short-term memory network, and bidirectional GRU prediction models, the proposed algorithm reduces the prediction errors of electric load by 0.55%, 0.49%, and 0.17%, respectively. The prediction errors of cooling load are reduced by 0.71%, 0.54%, and 0.19%, respectively. The prediction errors of heat load are reduced by 0.62%, 0.55%, and 0.19%, respectively. Conclusions The proposed algorithm model captures key information more effectively, demonstrates certain advantages over other algorithms, and improves prediction accuracy.

Key words: integrated energy systems, multivariate load prediction, energy coupling, composite correlation coefficient, encoder-decoder, gated recurrent unit, filtering

中图分类号: