基于ICEEMDAN-TCN-Transformer的短期光伏发电功率预测
Short-term photovoltaic power prediction based on ICEEMDAN-TCN-Transformer model
摘要
[目的]传统化石能源的使用引发环境污染等问题,因此光伏发电功率预测近年来受到广泛的关注.对此提出一种基于 ICEEMDAN 分解算法的短期光伏发电功率预测模型,该模型结合时间卷积网络与 Transformer 结构,旨在提升预测准确性.[方法]首先,采用 ICEEMDAN 算法对已有的光伏功率进行分解,分解光伏发电功率输入序列的高频波动和低频趋势,以降低其复杂度和随机波动性;并将时间卷积网络(TCN)与 Transformer 网络结合,通过残差连接和膨胀卷积的优势以及 Transformer的自注意力机制,显著提高光伏功率预测的准确性.通过澳大利亚Solibro数据集进行实验.[结果]对比实验表明,该文所采用的 ICEEMDAN-TCN-Transformer 光伏功率预测方法预测精度达到97.68%.[结论]该文可为实现高精度光伏功率预测提供一种可行且有效的技术方案.
[Objective]The use of traditional fossil energy has caused environmental pollution and other problems,so the power prediction of photovoltaic power generation has received extensive attention in recent years.In this paper,a short-term PV power prediction model based on the ICEEMDAN decomposition algorithm is proposed,which combines a temporal convolutional network and a Transformer structure to improve the prediction accuracy.[Methods]Firstly,the ICEEMDAN algorithm is used to decompose the existing photovoltaic power,and the high-frequency fluctuation and low-frequency trend of the photovoltaic power input sequence are decomposed to reduce its complexity and random fluctuation.By combining the time convolutional network(TCN)with the Transformer network,the accuracy of PV power prediction is significantly improved through the advantages of residual connection and dilated convolution,as well as the self-attention mechanism of Transformer.[Results]Experiments on the Australian Solibro dataset show that the prediction accuracy of the proposed ICEEMDAN-TCN-Transformer photovoltaic power prediction method reaches 97.68%.[Conclusion]This paper can provide a feasible and effective technical solution for achieving high-precision photovoltaic power forecasting.
年,卷(期):
2026, 52(9)
分类号:
TB9
TM46
机标分类号:
TM615
TP391.41
TM721
在线出版日期:
2026年10月14日
基金项目:
18JR3RA156:甘肃省自然科学基金项目2017-4-105:兰州科技计划项目,