中国农业气象 ›› 2022, Vol. 43 ›› Issue (01): 1-16.doi: 10.3969/j.issn.1000-6362.2022.01.001

• 农业生态环境栏目 •    下一篇

气象资料受限条件下BP神经网络优化模型模拟参考作物蒸散量:以京津冀地区为例

贾悦,苏永军,张冉,李鹏程,王凤春,路梅   

  1. 1.河北水利电力学院遥感与智慧水利创新中心,沧州 061001;2.四川大学水力学与山区河流开发保护国家重点实验室,成都 610065
  • 收稿日期:2021-05-06 出版日期:2022-01-20 发布日期:2022-01-15
  • 作者简介:苏永军,副教授,主要从事节水灌溉理论与技术研究,E-mail: 15130800077@163.com
  • 基金资助:
    河北省高等学校科学技术研究项目(QN2021227);河北省水利科研与推广计划项目(2020-64);沧州市重点研发计划指导项目(204107007)

Simulation of Reference Crop Evapotranspiration by BP Neural Network Optimization Model with Limited Meteorological Data: A Case Study in the Beijing-Tianjin-Hebei Region

JIA Yue, SU Yong-jun, ZHANG Ran, LI Peng-cheng, WANG Feng-chun, LU Mei   

  1. 1. Hebei University of Water Resource and Electric Engineering & Remote Sensing and Smart Water Innovation Center, Cangzhou 061001, China; 2. State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, Chengdu 610065
  • Received:2021-05-06 Online:2022-01-20 Published:2022-01-15

摘要: 为确定气象数据缺乏地区参考作物蒸散量(ET0)的最优简化估算模型,本文以京津冀地区作为研究区域,以传统BP神经网络模型为基础,基于粒子群算法(PSO)、遗传算法(GA)、思维进化算法(MEA)、麻雀算法(SSA)和人工鱼群算法(AF)5种优化算法,构建了PSO-BP、GA-BP、MEA-BP、SSA-BP、AF-BP共5种优化模型,并将计算结果与3种传统机器学习模型BP模型、随机森林模型(RF)、小波神经网络模型(WNN)和2种经验模型Hargreaves模型(HS)、Droogres-Allen模型(DA)进行对比,在仅输入温度数据的条件下,得出区域ET0最优估算模型。结果表明:在不同区域,5种优化模型计算精度显著高于其余模型,其中,SSA-BP模型均表现出了较高的精度,RMSE、R2、Ens和MAE分别为0.297~0.402mm·d−1、0.879~0.946、0.862~0.940、0.210~0.300mm·d−1,模型GPI在研究区域内排名第1位;在相同气象数据条件下,机器学习模型精度优于HS模型和DA模型,其中SSA-BP模型精度最高。因此,在仅有温度资料的条件下,SSA-BP模型可作为京津冀地区ET0估算的标准模型使用。

关键词: 京津冀地区, 参考作物蒸散量, BP神经网络模型, 麻雀算法

Abstract: In order to obtain the optimal simplified estimation model for reference crop evapotranspiration(ET0) in areas with a lack of meteorological data, the authors took the Beijing-Tianjin-Hebei region as the research area. Particle Swarm Algorithm(PSO), Genetic Algorithm(GA), Evolutionary Mind Algorithm(MEA), Sparrow Algorithm(SSA), and Artificial Fish Swarm Algorithm(AF) were used to optimize the BP model. Five optimization models: PSO-BP, GA-BP, MEA-BP, SSA-BP, AF-BP were constructed. These models were compared with BP model, random forest model(RF), wavelet neural network model(WNN), Hargreaves model(HS) and Droogrs-Allen model(DA). The results showed that: in different regions, the calculation accuracy of the five optimization models is significantly higher than the other models. SSA-BP model had the highest accuracy in different areas, with RMSE, R2, Ens and MAE of 0.297−0.402mm·d−1, 0.879−0.946, 0.862−0.940 and 0.210−0.300mm·d−1, respectively. The portability analysis results of the SSA-BP model showed that the model had strong generalization ability in the Beijing-Tianjin-Hebei region, and can achieve accurate estimation of ET0 from different stations. Thus, with only temperature data, the SSA-BP model can be used as a standard model for ET0 estimation in the Beijing- Tianjin-Hebei region.

Key words: Beijing-Tianjin-Hebei region, Reference crop evapotranspiration, BP neural network model, Sparrow algorithm