欢迎访问《太阳能学报》官方网站,今天是 分享到:
ISSN 0254-0096 CN 11-2082/K

太阳能学报 ›› 2022, Vol. 43 ›› Issue (10): 167-176.DOI: 10.19912/j.0254-0096.tynxb.2021-0453

• • 上一篇    下一篇

数据驱动的时间注意力卷积风电功率预测模型

张浩田1, 温蜜1, 李晋国1, 田英杰2   

  1. 1. 上海电力大学计算机科学与技术学院,上海 200090;
    2. 国网上海市电力公司电力科学研究院,上海 200437
  • 收稿日期:2021-04-25 出版日期:2022-10-28 发布日期:2023-04-28
  • 通讯作者: 温蜜(1979——),女,博士、教授、博士生导师,主要从事智能电网与电力大数据方面的研究。miwen@shiep.edu.cn
  • 基金资助:
    国家自然科学基金(61872230; 61802248; 61802249); 上海市2019年度“科技创新行动计划”高新技术领域项目(19511103700); 上海市科委项目(20020500600)

DATA DRIVEN TIME ATTENTION CONVOLUTION WIND POWER PREDICTION MODEL

Zhang Haotian1, Wen Mi1, Li Jinguo1, Tian Yingjie2   

  1. 1. School of Computer Science and Technology, Shanghai University of Electric Power, Shanghai 200090, China;
    2. Electric Power Research Institute of State Grid Shanghai Electric Power Company, Shanghai 200437, China
  • Received:2021-04-25 Online:2022-10-28 Published:2023-04-28

摘要: 由于风电受气象特征影响大,风能波动性和间歇性强,导致快速、精准的风电预测成为一个难题。对此,该文提出一种基于数据驱动的时间注意力卷积网络的风电功率预测方法。首先,将来自风力机和数据采集(SCADA)系统的数据进行清洗;然后采用可并行计算的时间卷积网络,并加入Attention机制突出关键特征的影响,使模型训练速度和预测精度得到有效提升。实验结果表明,该文所提方法与其他方法相比可更准确地减少数据噪声,同时有更高的预测精度和更快的训练速度。

关键词: 风力发电, 异常检测, 神经网络, 预测, 注意力机制

Abstract: Because wind power is greatly affected by meteorological characteristics, and ther wind energy has strong volatility and intermittence, so that fast and accurate wind power prediction becomes a difficult problem. Therefore, a data-driven time attention convolution network wind power prediction method is proposed. Firstly, the data from the wind turbine and SCADA system are cleaned. Then the Temporal convolutional network which can be calculated in parallel is adopted, and the attention mechanism is added to highlight the influence of key features, so that the training speed and prediction accuracy of the model are effectively improved. Experimental results show that compared with other methods, the proposed method can reduce data noise more accurately, and has higher prediction accuracy and faster training speed.

Key words: wind power, outlier detection, neural network, forecasting, attention mechanism

中图分类号: