中国农业气象 ›› 2016, Vol. 37 ›› Issue (06): 674-681.doi: 10.3969/j.issn.1000-6362.2016.06.007

• 论文 • 上一篇    下一篇

基于非线性PLSR模型的气候变化对粮食产量的影响分析

陈纪波,胡慧,陈克垚,王桂芝   

  1. 1.南京信息工程大学数学与统计学院,南京 210044; 2.中国气象局国家气候中心,北京 100081
  • 收稿日期:2016-03-20 出版日期:2016-12-20 发布日期:2016-12-15
  • 作者简介:陈纪波(1961-),副教授,研究方向为应用统计。E-mail:chenjibo@nuist.edu.cn
  • 基金资助:
    国家社会科学基金(15BTJ019)

Effects of Climate Change on the Grain Yield Based on Nonlinear PLSR Model

CHEN Ji-bo,HU Hui ,CHEN Ke-yao ,WANG Gui-zhi   

  1. 1.School of Mathematics and Statistics, Nanjing University of Information Science and Technology,Nanjing 210044, China; 2. National Climate Center, China Meteorological Administration, Beijing 100081
  • Received:2016-03-20 Online:2016-12-20 Published:2016-12-15

摘要: 考虑气候因子间多重共线性及其与粮食产量间复杂的非线性关系,本文在HP滤波分离出气候产量的基础上,尝试引入基于三次B样条变换(Spline-PLSR)和内部嵌入GRNN的两种非线性偏最小二乘模型(GRNN-PLSR),利用1961-2008年气候因子数据建立气候产量计算模型,以2009—2013年数据进行拟合检验,并与常用的C-D生产函数法计算的气候产量进行比较。结果表明,Spline-PLSR法在拟合气候因子变化对粮食产量影响时预测精度较高。而且,与C-D生产函数法相比,Spline-PLSR所需要素较少,操作简单,相对误差最高仅为13.6%;与GRNN-PLSR法拟合结果相比,Spline-PLSR相对误差波动较小,因此,基于三次B样条变换的非线性偏最小二乘法建模较适合拟合气候产量。

关键词: 气候产量, 偏最小二乘法, 三次B样条, 广义回归神经网络

Abstract: Considering the multicollinearity of climatic factors,as well as the complex nonlinear relationship between climatic factors and the grain yield, authors attempt to model the climatic factors and climate yield data from 1961 and 2008 in this paper with respect to the cubic B splines function(Spline-PLSR)and internal embedded Generalized regression neural network(GRNN) into the partial least squares regression,on the basis of separating the climatic yield by HP filter. Through?the fitting?test based on the data from 2009 to 2013 and the comparison between the C-D production function and the proposed model,authors determine that the Spline-PLSR model is relatively simple with higher prediction accuracy. Compared with the C-D production function,the Spline-PLSR model requires fewer elements and possesses a better forecasting value. It is worth noting that the fitting result of Spline-PLSR is more stable than that of GRNN-PLSR. Hence,it is a better choice to utilize Spline-PLSR to fit the influence of climatic factors on the grain yield.

Key words: Climatic yield, Partial least squares regression, Cubic B-spline, Generalized regression neural network