基于小波包变换与深度学习的超短期光伏功率预测
ULTRA-SHORT-TERM PHOTOVOLTAIC POWER FORECASTING BASED ON WAVELET PACKET TRANSFORM AND DEEP LEARNING
刘源延 1孔小兵 1马乐乐 1刘向杰1
作者信息
- 1. 华北电力大学控制与计算机工程学院,北京 102206
- 折叠
摘要
针对光伏功率序列的复杂多变特征,提出一种基于小波包变换(WPT)的门控循环单元(GRU)光伏功率组合预测方法.首先通过相关性分析挑选重要气象因子,并利用WPT将原始光伏功率序列分解为一组子序列;然后,提出一种基于莱维飞行天牛须搜索算法(LFBAS)的相似日选择方法,以选择相似于预测日的历史日作为输入数据集;最后,建立一组基于GRU网络的深度学习光伏功率预测模型,将每个子序列预测结果叠加得到光伏功率最终预测结果.仿真结果表明,该文所提出的预测方法在预测精度和计算效率方面具有显著优势.
Abstract
Considering the highly varying and complex features of photovoltaic power generation,this paper constitutes a hybrid Photovoltaic(PV)power forecasting(PVPF)method based on gated recurrent unit(GRU)combining with wavelet packet transform(WPT)algorithm.First,correlation analysis is used to select the main meteorological factors while wavelet packet decomposition is used to decompose the original PV power into a series of sub-signals.A similar day selection method based on levy-flight BAS algorithm is proposed to select historical days similar to the forecast day from the real-time massive data.Deep learning model for PVPF is established using a group of GRU networks.These GRU forecasting sub-signals are synthesized to form the final forecasting PV power.The simulation results verify that the proposed method shows obvious advantages in terms of both forecasting accuracy and computational efficiency.
关键词
光伏发电/功率预测/小波包变换/相似日/门控循环单元/天牛须搜索算法Key words
PV power/power forecasting/wavelet packet transform/similar day/gated recurrent unit/beetle antennae search algorithm引用本文复制引用
基金项目
国家重点研发计划(2021YFE0190900)
国家自然科学基金(62073136)
国家自然科学基金(61833011)
国家自然科学基金(62203170)
出版年
2024
NETL
NSTL
万方数据