应用盲分离神经网络预测逐日太阳辐射能  

APPLICATION ON FORECAST OF DAY-BY-DAY SOLAR IRRADIANCE BASED ON BLIND SOURCE SEPARATION NEURAL NETWORK

作　　者：周勃[1] 陈长征[1] 宋楠 

机构地区：[1]沈阳工业大学建筑工程学院,沈阳110023 [2]辽宁省城乡规划设计研究院,沈阳110015

出　　处：《太阳能学报》2007年第9期1008-1011,共4页Acta Energiae Solaris Sinica

基　　金：辽宁省自然科学基金(20052042);辽宁省教育厅高等学校科学研究项目(05L284)

摘　　要：提出一种应用盲分离神经网络预测逐日太阳辐射能的方法。首先在卷积混合基础上,应用最大化负熵准则分离原始太阳辐射时间序列,从观测数据中剔除不可靠信息;考虑到太阳负荷的特点,将分离后的样本输入到径向基函数神经网络(RBFN)中,通过调整参数训练网络直到满足约束条件为止,由此恢复盲分离所带来的幅值和排列顺序变化;最后分别比较盲分离神经网络、RBFN和BP网络的预测误差值,结果说明本文建立的模型提高了预测的准确度。A method of forecasting the total solar irradiance based on blind source separation （BSS） neural network was presented. First, we used this method to separate the initial time sequence of day-by-day solar irradiance to eliminate the unreliable information. In consideration of the complex behavior of solar irradiance, either periodic or random, a kind of dynamic neural network, radial basis function neural network RBFN, was used for such case. After that the separating results were supplied to the input layer and were trained through adjusting the number of neurons and the weights in different layers of the network until the errors reached the stop conditions. Finally the forecasting model mentioned in this paper was tested through a practical sample, which indicates that the accuracy of the model is mere satisfactory than without blind source separation. Thus the method proposed in this paper can also be applicable to new energy and other relating fields.

关 键 词：太阳辐射 盲分离 神经网络 径向基函数 

分 类 号：TK519[动力工程及工程热物理—热能工程]