摘要
宿州春季严重干旱序列数据偏少,可用传统GM(1,1)模型进行预测,但由于序列变化幅度较大,预测效果不理想。本文利用灰色与BP神经网络组合模型对宿州春季重旱发生年份进行预测,即首先弱化序列变化幅度,并改进GM(1,1)模型导数信息处理方式,构建可逼近精度目标的m-GM(1,1)预测模型,然后应用BP神经网络对m-GM(1,1)模型的残差进行拟合,对m-GM(1,1)预测模型进行修正。结果表明,灰色神经网络组合模型的精度(|Q|=0.0045)比单一的1.7-GM(1,1)模型(|Q|=4.18)和传统的单一GM(1,1)模型精度(|Q|=9.36)提高许多。预测2005年后的下一个宿州市春季严重干旱发生年份为2009年,可以作为预报当地春季干旱时的参考,并结合其他方法作进一步预测,为当地防灾减灾提供科学依据。
The traditional GM （ 1,1 ） model can be employed to forecast spring drought in Suzhou with small amount of data, but this model is not ideal due to the large scope of sequence changes. The gray and BP neural network model were adopted to predict the years of the serious spring drought occurrence in Suzhou. The scope of data sequence was weakened and the disposal of differential coefficient of GM（ 1,1 ） model was improved to build near - precision model m-GM （ 1,1 ） and to revise the residual error of m-GM （ 1,1 ） model. The result showed that the precision of the gray neural network model （ ｜Q｜ = 0. 0045 ） was much higher than that of the single 1.7-GM （ 1,1 ） model （ ｜Q｜ = 4.18 ） and the traditional single GM （ 1,1 ） model （ ｜Q｜ = 9.36）. The year of 2009 would be the next serious spring drought year after 2005 forecasted by the model.
出处
《中国农业气象》
CSCD
2009年第2期271-274,共4页
Chinese Journal of Agrometeorology
