欢迎访问《农学学报》，
农学学报

农学学报 ›› 2023, Vol. 13 ›› Issue (8): 89-93.doi: 10.11923/j.issn.2095-4050.cjas2023-0009

• 农业信息 农业气象 • 上一篇    下一篇

基于灰色关联下神经网络在南果梨始花期预报中的应用

战莘晔1(), 吕晓2, 金丹丹3(), 康晓玉1, 韩艳凤1, 韩国敬1, 徐庆喆1, 田璐1   

  1. 1 鞍山市气象局，辽宁鞍山 114004
    2 锦州市生态与农业气象中心，辽宁锦州 121000
    3 辽宁省农业科学院植物营养与环境资源研究所，沈阳 110161
  • 收稿日期:2022-12-27 修回日期:2023-01-30 出版日期:2023-08-20 发布日期:2023-08-16
  • 通讯作者: 金丹丹，女，1985年出生，辽宁鞍山人，副研究员，博士，研究方向：农业资源利用。通信地址：110161 辽宁省沈阳市沈河区东陵路84号 辽宁省农业科学院植物营养与环境资源研究所，Tel：024-31028698，E-mail：61985155@qq.com。
  • 作者简介:

    战莘晔，女，1989年出生，辽宁鞍山人，工程师，硕士，研究方向：农业气象。通信地址：114004 辽宁省鞍山市常青街16号 鞍山市气象局，Tel：0412-5835554，E-mail：。

  • 基金资助:
    辽宁省气象局科研课题计划“鞍山南果梨花期预测”(2019SXG11); 黑土地保护与利用科技创新工程专项资助“稻作农田土壤地力保持与提质增效模式”(XDA28090300)

Application of Neural Networks Based on Gray Relation in Forecasting the Initial Flowering Date of Nanguo Pear

ZHAN Shenye1(), LV Xiao2, JIN Dandan3(), KANG Xiaoyu1, HAN Yanfeng1, HAN Guojing1, XU Qingzhe1, TIAN Lu1   

  1. 1 Anshan Meteorological Administration, Anshan 114004, Liaoning, China
    2 Jinzhou Ecological and Agricultural Meteorological Center, Jinzhou 121000, Liaoning, China
    3 Plant Nutrition and Environmental Resources Research Institute, Liaoning Academy of Agricultural Sciences, Shenyang 110161, Liaoning, China
  • Received:2022-12-27 Revised:2023-01-30 Online:2023-08-20 Published:2023-08-16

RichHTML

1

PDF (PC)

59

摘要：

为探寻更精确有效的南果梨始花期预报方法，采用灰色关联分析法确定与始花期关联较大的冬季气象因子，以此作为BP(Black Propagation)神经网络与RBF(Radial Basis Function)神经网络建模的输入因子并预测南果梨始花期，利用均方根误差(RMSE)和相对误差(RE)评价该模型的预测效果，同时对比与南果梨始花期显著相关的冬季气象因子建立逐步回归方程并进行回代后的预测结果。结果表明：（1）与南果梨始花期灰色关联较大的气象因子为冬季日均气温、日最高气温、日最低气温、日均相对湿度，关联度均在0.6以上，故将这4个因子作为BP和RBF神经网络模型的输入层来预测南果梨始花期；（2）与始花期显著相关的气象因子有日均气温、日均5 cm地温、日最低气温、日最高气温，相关系数分别为-0.646、-0.628、-0.638、-0.663，所建回归模型均通过了显著性检验且具有统计学意义；（3）BP和RBF方法建立的模型拟合精度总体上较接近；（4）基于灰色关联下BP神经网络和RBF神经网络预测结果误差分别为1 d和2.25 d，BP神经网络预测的开花日期更接近实际开花日期；（5）基于灰色关联下BP神经网络模型RMSE为1、RE为6.34%、R2为0.7，而RBF神经网络模型RMSE为2.25、RE为13.13%、R2为0.568。综上，灰色关联分析法建立的BP神经网络模型较RBF模型预测南果梨始花期更精确。

关键词: 始花期, 灰色关联分析法, BP神经网络, RBF神经网络, 逐步回归分析

Abstract:

To explore a more accurate and effective method to predict the initial flowering date of Nanguo pear, grey relation analysis was used to determine the winter meteorological factors associated with the initial flowering date, which could be used as BP (Black Propagation) neural network and RBF (Radial Basis Function). The root mean square error (RMSE) and relative error (RE) were used to evaluate the prediction effect of the model. Meanwhile, the stepwise regression equation was established by comparing the winter meteorological factors significantly related to the initial flowering date of Nanguo pear and the prediction results after back substitution. The following are the results. (1) The meteorological factors with greater grey relation with the initial flowering date of Nanguo pear were the daily average temperature, maximum temperature, minimum temperature and relative humidity in winter, and the correlation degree was above 0.6. Therefore, these four factors were used as the input layer of BP and RBF neural network models to predict the initial flowering date of Nanguo pear. (2) The meteorological factors significantly correlated with the onset of flowering were daily mean temperature, daily mean ground temperature of 5 cm, daily minimum temperature and daily maximum temperature, and the correlation coefficients were -0.646, -0.628, -0.638 and -0.663, respectively. All the established regression models passed the significance test and had statistical significance. (3) The accuracy of model fitting established by BP and RBF was generally close. (4) The errors of BP neural network and RBF neural network prediction results based on grey relation were 1 day and 2.25 days respectively, and the flowering date predicted by BP neural network was closer to the actual flowering date. (5) RMSE, RE and R2 of BP neural network model based on grey relation were 1, 6.34% and 0.7, respectively, and for RBF neural network model, RMSE was 2.25, RE was 13.13%, and R2 was 0.568. In conclusion, BP neural network model established by grey relation analysis method was more accurate than RBF model in predicting the initial flowering date of Nanguo pear.

Key words: initial flowering date, grey relation analysis, BP neural network, RBF neural network, stepwise regression analysis