Scholarlink

The addition of energy storage can enhance the dispatch capability of PV power plants and help them better adapt to grid demand. One of the keys to the stable control of a combined photovoltaic and storage power system is the accurate prediction of PV power generation. This paper proposes an FFTformer predictive model based on Fast Fourier Transform (FFT), 1-D CNN-LSTM and multi-timescale information. Firstly, the proposed model uses Fourier Transform to remove the noise of different timescale data. Secondly, the multi-attention mechanism is employed to strengthen the characterization ability of the data. At the same time, power features at different time scales are extracted using 1-D CNN networks and LSTM networks. Finally, the multi-timescale features extracted by the attention mechanism and 1D CNN-LSTM are used to predict the photovoltaic (PV) power. To validate the effectiveness of the model, a historical dataset from Hebei Province, China is used for various types of experiments. Among them, the effectiveness and advantages of the proposed model FFORMER are verified by comparison tests with other models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-Term Photovoltaic Power Forecasting Using Multi-timescale Information Based FFTformer Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Haohao Feng,
  • Xiaoyun Niu,
  • Mifeng Ren,
  • Junghui Chen,
  • Wenjie Zhang

摘要

The addition of energy storage can enhance the dispatch capability of PV power plants and help them better adapt to grid demand. One of the keys to the stable control of a combined photovoltaic and storage power system is the accurate prediction of PV power generation. This paper proposes an FFTformer predictive model based on Fast Fourier Transform (FFT), 1-D CNN-LSTM and multi-timescale information. Firstly, the proposed model uses Fourier Transform to remove the noise of different timescale data. Secondly, the multi-attention mechanism is employed to strengthen the characterization ability of the data. At the same time, power features at different time scales are extracted using 1-D CNN networks and LSTM networks. Finally, the multi-timescale features extracted by the attention mechanism and 1D CNN-LSTM are used to predict the photovoltaic (PV) power. To validate the effectiveness of the model, a historical dataset from Hebei Province, China is used for various types of experiments. Among them, the effectiveness and advantages of the proposed model FFORMER are verified by comparison tests with other models.