中国农业气象 ›› 2019, Vol. 40 ›› Issue (04): 240-249.doi: 10.3969/j.issn.1000-6362.2019.04.005

• 论文 • 上一篇    下一篇

南方塑料大棚“棚温逆差”特征及其预报

杨栋,丁烨毅,孙军波,李清斌,魏莎莎,黄鹤楼   

  1. 1.宁波市气象台,宁波 315012;2.慈溪市气象局,宁波 315033
  • 出版日期:2019-04-20 发布日期:2019-04-17
  • 作者简介:杨栋(1988-),工程师,硕士,主要从事气候变化与农业气象研究。E-mail: yangdong_314@163.com
  • 基金资助:

    宁波市气象局一般项目(NBQX2017004B) ;浙江省重点研发计划(2018C02011)

Characteristics and Forecasting Technology of Greenhouse Temperature Deficit in Southern Plastic Greenhouse

YANG Dong, DING Ye-yi, SUN Jun-bo, LI Qing-bin, WEI Sha-sha, HUANG He-lou   

  1. 1. Ningbo Bureau of Meteorology, Ningbo 315012, China; 2. Cixi Bureau of Meteorology, Ningbo 315033
  • Online:2019-04-20 Published:2019-04-17

摘要:

“棚温逆差”指设施大棚采取保温措施时出现棚内日最低气温低于棚外的现象,利用2010-2015年冬季和初春浙江地区塑料大棚内外气象资料,结合大棚覆膜保温、开窗通风等人工操作记录,对南方塑料大棚“棚温逆差”发生的特征及关键影响因子进行探究;选用5种常见神经网络方法(BP、GA-BP、RBF、GRNN、PNN)分别构建“棚温逆差”预报模型,并基于不同模型预报准确率构建集合预报模型。结果表明:(1)初春和初冬季“棚温逆差”频率较严冬高3倍,棚内1.5m高处出现概率较0.5m处偏高3倍;0.5m高处大棚边缘出现“棚温逆差”概率为中央的8~13倍,1.5m高处中央和边缘位置出现概率差异较小。(2)“棚温逆差”发生时,棚外日最低气温主要在2~11℃区间,大棚保温方式为单膜或双膜覆盖,其中单膜覆盖占比达93%以上。“棚温逆差”多发生在白天东西侧窗开启高度较高(平均高度35~40cm)且夜间放小缝通风时;0.5m处发生“棚温逆差”时,天气条件较1.5m处发生时明显偏差,日最小相对湿度、总云量和日均风速偏高,日照时数偏少。(3)5种神经网络方法对“棚温逆差”的预报准确率在80%左右,其中GA-BP模型预报准确率最高;集合预报模型对0.5m高度处“棚温逆差”预报准确率在85%左右,1.5m高度处准确率在80%左右,且预报稳定性较单一模型高。

关键词: 棚温逆差, 塑料大棚, 温度, 神经网络, 集合模型

Abstract:

The Greenhouse Temperature Deficit (GTD) refers to the phenomenon that the daily minimum temperature in the greenhouse is lower than that outside the greenhouse when insulation measures have been taken. By using the meteorological data and manual operation records (film-covered insulation, window ventilation and so on) of plastic greenhouses in Zhejiang province during the winter and early spring of 2010-2015, the characteristics and key influencing factors of the GTD in southern China were explored. Five common neural networks (BP, GA-BP, RBF, GRNN, PNN) were employed to construct the GTD prediction models. The ensemble forecasting model of the result was built based on the forecasting accuracy of the selected five models. The results showed that: (1) the frequency of GTD in early spring and winter (Mar. and Dec.) was three times higher than that in severe winter (Jan.-Feb). The occurrence probability of the GTD at 1.5m was three times higher than that at 0.5m. At the height of 0.5m, the probability of GTD at the edge was 8 to 13 times higher than that at the center. At the height of 1.5m, the difference in the probability of GTD between the center and edge was small. (2) When the GTD happened, the daily minimum temperature outside the greenhouse mainly concentrated on 2 to 11 degrees Celsius, and the insulation method was single film or double film coverage, the proportion of single film coverage was over 93%. The GTD mostly occurred when the opening height of the east and west side windows was high (35?40cm) in daytime and the small gap ventilation was put in the night. When the GTD occurred at 0.5m, the weather conditions were obviously deviated from that of 1.5m. The daily minimum relative humidity, total cloud and average daily wind speed were higher, and the sunshine hours were less. (3) The prediction accuracy of the five selected neural networks for the GTD was basically about 80%, and the GA-BP model was with the highest prediction accuracy. The ensemble model for GTD at 0.5m height was above 85%, and that at 1.5m height was above 80%. The stability of ensemble model was better than that of single model.

Key words: Greenhouse temperature deficit, Plastic greenhouse, Temperature, Neural network, Ensemble prediction