发电技术 ›› 2021, Vol. 42 ›› Issue (1): 60-68.DOI: 10.12096/j.2096-4528.pgt.20103

• 电力系统规划 • 上一篇    下一篇

碳中和背景下多通道特征组合超短期风电功率预测

黄树帮1,2(), 陈耀1,2(), 金宇清3,*()   

  1. 1 新疆金风科技股份有限公司, 北京市 大兴区 100176
    2 江苏金风软件技术有限公司, 江苏省 无锡市 214000
    3 河海大学能源与电气学院, 江苏省 南京市 211100
  • 收稿日期:2020-10-09 出版日期:2021-02-28 发布日期:2021-03-12
  • 通讯作者: 金宇清
  • 作者简介:金宇清(1980),男,博士,副教授,研究方向为新能源发电系统的建模与控制,本文通信作者,jyq16@hhu.edu.cn
    黄树帮(1979),男,硕士,高级工程师,研究方向为智能变电站及新能源发电并网技术,huangshubang@goldwind.com.cn
    陈耀(1990),男,硕士,工程师,研究方向为人工智能与功率预测,yao_chen@aliyun.com
  • 基金资助:
    国家自然科学基金项目(52077059)

A Multi-channel Feature Combination Model for Ultra-short-term Wind Power Prediction Under Carbon Neutral Background

Shubang HUANG1,2(), Yao CHEN1,2(), Yuqing JIN3,*()   

  1. 1 Xinjiang Goldwind Science & Technology Co., Ltd., Daxing District, Beijing 100176, China
    2 Jiangsu Goldwind Software & Technology Co., Ltd., Wuxi 214000, Jiangsu Province, China
    3 College of Energy and Electrical Engineering, Hohai University, Nanjing 211100, Jiangsu Province, China
  • Received:2020-10-09 Published:2021-02-28 Online:2021-03-12
  • Contact: Yuqing JIN
  • Supported by:
    National Natural Science Foundation of China(52077059)

摘要:

碳中和背景下,风电将成为我国的主导能源之一。随着人工智能技术快速发展,人工神经网络被广泛应用于风力发电功率预测。传统的人工神经网络算法采用固定形式数据集和单一网络结构,限制了整体表达能力,导致超短期风电功率预测由于各种不确定因素造成难以控制的误差。为此,提出一种基于人工神经网络的多通道特征组合模型,用于超短期风电功率预测。首先将数据集进行重新分类,分别输入到3个神经网络,建立3种特征组合形式;再将多通道特征进行拼接融合,并将融合后的特征加入到全连接神经网络中进行功率预测,可消除不同特征之间的干扰,有效学习到长期依赖的数据特征;最后对5个风电场数据进行算法验证。实验结果表明,该方法比单通道模型能够获得更好的预测精度,而且增加了网络稳定性。

关键词: 碳中和, 风力发电, 超短期功率预测, 人工神经网络, 多通道

Abstract:

Wind power will become one of the dominant power sources of China oriented to carbon neutral. With the rapid development of artificial intelligence technology, artificial neural networks are widely used in wind power generation forecasting. Traditional artificial neural network algorithms use fixed-form data sets and simple network structures, which limits the overall expression ability and results in uncontrollable errors in ultra-short-term wind power forecasting due to various uncertain factors. In this work, a multi-channel feature combination model based on artificial neural network for ultra-short-term wind power prediction was proposed. Firstly, the data were reclassified and input into three neural networks to establish three feature combinations. After that, multi-channel features splicing and fusion were performed. The fused features were added to the fully connected neural network for power prediction, which can eliminate the interference between different features and effectively learn long-term dependent data features. Finally, the algorithm was verified on the actual data of five wind farms. The experimental results show that this method has better prediction accuracy than the single-channel model, and can improve the network stability.

Key words: carbon neutral, wind power generation, ultra-short-term power prediction, artificial neural network, multi-channel

中图分类号: