摘要
针对于传统的确定性太阳辐射模型不能反映气象变化的弊端,提出了基于回归BP神经网络和小波分析理论的太阳散射辐射逐日预测模型。神经网络具有非线性函数逼近及自组织自学习的能力,基于小波分析在信号处理方面的时频域多分辨特性,本文利用小波变换将太阳散射辐射数据序列进行时频域分解后作为神经网络预测模型的输入样本,实例表明该方法与传统模型相比预测精度高,具有可行性。
Classic solar radiation models are meteorological variety irrespectively. In order to avoid this disadvantage, a daily solar diffuse radiation forecasting model based on recurrent BP network and wavelet analysis is put forward. Artficial neural network has the capacity of non-linear function approximation, self-orgnizing and self-learning. Based on multi-resolution advantadge of wavelet analysis＇s time-frequency domains, the wavelet transform method is adopted to decompose the solar diffuse radiation sequence into various time-frequency domains as the input data of the artficial neural network. An example indicates the accuracy of this method is higher than that of the methods reported before, and has higher feasibility.
出处
《建筑热能通风空调》
2006年第6期76-79,共4页
Building Energy & Environment
关键词
太阳散射辐射
递归BP网络
小波变换
相关系数
solar diffuse radiation, recurrent BP network, wavelet transform, correlation coefficient
