小波分析在太阳辐射神经网络预测中的应用研究

Study of Application of Wavelet Analysis to Neural Networks for the Forecast of Solar Irradiance

通过小波变换将太阳辐射数据序列分解到不同的时频域上 ,并对每一频域分量建立一个递归BP网络模型 ;然后用网络模型对各频域分量进行预测 ,将各预测结果进行代数叠加 ,从而得到太阳辐射的预测结果。为体现近期预测结果在精度上的相对重要性 ,在递归BP网络的权阈值修改算法中 ,引入了折扣系数法。最后 ,通过对上海太阳日总辐射的预测实例表明 ,该方法在预测太阳辐射时是可行的。

In this paper, the data series of solar irradiance is mapped into several time-frequency ranges using wavelet transform, and a recurrent BP network is established for each frequency range. The solar irradiance can be predicted with the algebraic sum of the irradiance of each frequency range forecasted by the established network model. The discount coefficient method is adopted in modification of the weights and thresholds of the networks so as to make the closest forecast playing a more important role. An example is presented with the daily forecasted total solar irradiance in Shanghai, and the results indicate that the method is satisfactory for the forecast of solar irradiance.