计算机工程与应用 ›› 2024, Vol. 60 ›› Issue (9): 317-325.DOI: 10.3778/j.issn.1002-8331.2301-0155

• 工程与应用 • 上一篇    下一篇

针对光伏发电功率预测的LSTformer模型

刘世鹏,宁德军,马崛   

  1. 1.中国科学院 上海高等研究院,上海 200120
    2.中国科学院大学,北京 100049
  • 出版日期:2024-05-01 发布日期:2024-04-29

LSTformer Model for Photovoltaic Power Prediction

LIU Shipeng, NING Dejun, MA Jue   

  1. 1.Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 200120, China
    2.University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2024-05-01 Published:2024-04-29

摘要: 为了提高光伏发电功率预测精度,提出了一种基于长短期时序数据融合的Transformer生成式预测模型:LSTformer,能准确有效地预测光伏发电功率。LSTformer创新性地提出了时序分析模块(time series analysis,TSA)、时序特征融合模块(time series feature fusion,TSFF)和多周期嵌入模块(cycleEmbed),利用数据融合解决难以提取多时间尺度时序特征问题。设计时间卷积前馈(time convolution feedforward,TCNforward)单元,在编解码的过程中进一步提取时序特征。利用某光伏电站实际历史发电数据,通过实验验证LSTformer模型在光伏发电功率预测领域得到最低的均方误差(mean squared error,MSE)、平均绝对误差(mean absolute error,MAE),并通过消融实验验证了各模块的有效性。

关键词: Transformer, 长短期记忆网络, 跳跃-门控循环单元, 光伏发电功率预测, 时序数据预测

Abstract: In order to improve the prediction accuracy of photovoltaic power, a Transformer generation prediction model based on data fusion is proposed:LSTformer, which can accurately and effectively predict short-term photovoltaic power. LSTformer proposes a time series analysis (TSA) module, a time series feature fusion (TSFF) module and a cycleEmbed module. It uses data fusion to solve the problem that it is difficult to extract multiple time series features. The time convolution feedforward (TCNforward) unit is designed to further extract the timing characteristics in the process of encoding and decoding. Using the actual historical power data of a photovoltaic power station, the LSTformer model is verified to have the lowest mean squared error (MSE) and mean absolute error (MAE) in the field of photovoltaic power prediction through experiments, and the effectiveness of each module is verified through ablation experiments.

Key words: Transformer, long short term memory (LSTM), skip gated recurrent unit (skip-GRU), photovoltaic power prediction, time series data prediction