摘要
针对光伏功率时间序列的非平稳特性,提出一种基于最大重叠离散小波变换(MODWT)和长短期记忆网络(LSTM)的光伏功率组合预测模型。利用皮尔逊相关系数确定影响光伏功率的重要气象因素,基于MODWT算法对历史光伏功率序列进行分解,将选取的气象因素与分解得到的平稳子序列共同构成各个LSTM网络输入,通过汇总重构每个LSTM网络的子序列预测结果得到最终的光伏功率预测结果。从理论层面分析所建立的MODWT算法的完全重构性,并基于李雅普诺夫稳定性定理推导保证预测网络收敛的学习率范围。仿真对比结果显示,所提出的光伏功率预测模型在预测精度和鲁棒性方面具有明显优势。
Aiming at the non-stationary characteristics of PV power time series,this paper proposes a hybrid PV power forecasting model based on maximum overlap discrete wavelet transform(MODWT)and long short-term memory network(LSTM).First,Pearson correlation coefficient is used to identify important meteorological factors while MODWT is used to decompose the historical PV power series.The selected meteorological factors and the decomposed stationary subsequences are combined to form the input of each LSTM network.The sub-sequence prediction results of each LSTM network are integrated and reconstructed to the final PV power prediction results.The complete reconstruction of MODWT algorithm established in this paper is analyzed at the theoretical level,and the range of learning rate to ensure the convergence of the prediction network is derived based on Lyapunov stability theorem.The simulation results show that this proposed forecasting model has the obvious advantages in forecasting accuracy and robustness.
作者
马乐乐
孔小兵
郭磊
刘源延
刘向杰
Ma Lele;Kong Xiaobing;Guo Lei;Liu Yuanyan;Liu Xiangjie(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2024年第5期576-583,共8页
Acta Energiae Solaris Sinica
基金
国家重点研发计划(2021YFE0190900)
国家自然科学基金(62073136,62203170)
中国博士后科学基金(2022T150210)。
关键词
光伏功率预测
长短期记忆网络
非平稳时间序列分解
预测网络收敛性
photovoltaic power forecasting
long short-term memory network
non-stationary time series decomposition
convergence of prediction network
