摘要
为进一步提高超短期光伏功率预测的准确性,提出一种基于多模态多尺度特征的超短期光伏功率预测方法。首先,基于不同采样间隔得到多尺度地基云图与多尺度光伏功率作为预测模型的输入数据;其次,利用自注意卷积长短期记忆神经网络与长短期记忆神经网络分别提取多尺度云图数据的时空特征与多尺度功率数据的时序特征,从而得到多模态多尺度特征;然后,提出一种基于多头自注意力与多头交叉注意力的融合注意力机制,对多模态多尺度特征信息进行深度融合;最后,将多模态多尺度融合特征作为多层感知器的输入,进而实现超短期光伏功率预测。实验结果表明,该方法能够有效提高超短期光伏功率预测准确性。
To further improve the accuracy of ultra-short-term photovoltaic power forecasting,a forecasting method based on multimodal multi-scale features is proposed.Firstly,multi-scale historical cloud images and PV power are obtained based on different sampling intervals as the input data of the forecasting model.Secondly,the spatial and temporal features of multi-scale cloud image data and multi-scale power data are extracted using self-attention convolutional long short-term memory neural network and long short-term memory neural network respectively,to obtain multi-modal multi-scale features.Then,a fusion-attention mechanism based on multihead self-attention and multi-head cross-attention is proposed to integrate multi-modal multi-scale feature information deeply.Finally,the multi-modal multi-scale fusion features are used as the input of multi-layer perceptron to achieve ultra-short-term photovoltaic power forecasting.Experimental results show that this method can effectively improve the accuracy of ultra-short-term photovoltaic power forecasting.
作者
陈殿昊
臧海祥
刘璟璇
张越
孙国强
卫志农
Chen Dianhao;Zang Haixiang;Liu Jingxuan;Zhang Yue;Sun Guoqiang;Wei Zhinong(School of Electrical and Power Engineering,Hohai University,Nanjing 211100,China)
出处
《太阳能学报》
北大核心
2025年第8期472-480,共9页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(52077062)。
关键词
光伏功率
预测
深度学习
多模态数据
多尺度特征
注意力机制
photovoltaic power
forecasting
deep learning
multi-modal data
multi-scale features
attention mechanism
作者简介
通信作者:臧海祥(1986-),男,博士、教授,主要从事人工智能在电力系统中应用等方面的研究。zanghaixiang@hhu.edu.cn。
