摘要
针对光伏功率的波动性、随机性、间歇性,提出一种基于最大重叠小波变换(MODWT)、自适应噪声完备集合经验模态分解(CEEMDAN)、长短期记忆网络(LSTM)的光伏功率短期区间预测模型。首先利用MODWT和CEEMDAN将光伏功率时间序列进行二次分解得到本征模态函数(IMF)分量;再将这些IMF分量分别输入进LSTM进行分量预测并将分量预测结果重构得到点预测结果;最后利用分位数回归对点预测结果进行建模后得到区间预测结果。实际算例表明,时频域分解方法与频域分解方法的结合,使得该模型在3种天气情况下的光伏功率点预测和区间预测均表现出优异的鲁棒性和准确性。
In response to the volatility,randomness,and intermittency of photovoltaic power,this study proposes a short-term interval prediction model based on the maximum overlapping discrete wavelet transform(MODWT),complementary empirical mode decomposition with adaptive noise(CEEMDAN),and long short-term memory network(LSTM).Firstly,MODWT and CEEMDAN are used to decompose the PV power time series quadratically to obtain the intrinsic mode functions(IMF)components,Then,these IMF components are inputted into the LSTM for component prediction and the component prediction results are reconstructed to obtain the point prediction result;Finally,quantile regression is used to model the point prediction results and the interval prediction results are obtained.Finally,the interval prediction results are obtained by modeling the point prediction results with quantile regression.Practical examples show that the combination of the time-frequency domain decomposition method and the frequency domain decomposition method makes the model show excellent robustness and accuracy in the point prediction and interval prediction of PV power under the three weather conditions.
作者
陈船宇
熊国江
方厚康
罗颖勋
Chen Chuanyu;Xiong Guojiang;Fang Houkang;Luo Yingxun(College of Electrical Engineering,Guizhou University,Guiyang 550025,China;Guizhou University Survey,Design and Research Institute Co.,Ltd.,Guiyang 550025,China)
出处
《太阳能学报》
北大核心
2025年第2期416-424,共9页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(52167007
52367006)
贵州大学勘察设计研究院有限责任公司创新基金(贵大勘察[2022]03号)。
关键词
光伏功率
预测
深度学习
长短期记忆
最大重叠小波变换
自适应噪声完备集合经验模态分解
PV power
prediction
deep learning
long short-term memory
maximal overlap discrete wavelet transform
complete ensemble empirical mode decomposition with adaptive noise
