基于BP神经网络的冬季日光温室小气候模拟  

Microclimate Simulation of Sunlight Greenhouse in Winter Based on BP Neural Network

作　　者：王春玲[1,2,3] 魏瑞江[2,3] 申双和[1] 王鑫[2,3] 邢文发[2,3] 朱慧钦 敖雪[5] 

机构地区：[1]南京信息工程大学应用气象学院,南京210044 [2]河北省气象科学研究所,石家庄050021 [3]河北省气象与生态环境重点实验室,石家庄050021 [4]河北省高邑县气象局,河北高邑051330 [5]沈阳区域气候中心,沈阳110016

出　　处：《中国农学通报》2014年第5期149-157,共9页Chinese Agricultural Science Bulletin

基　　金：公益性行业(气象)科研专项"设施农业气象灾害预警及防御关键技术"(GYHY201006028);公益性行业(气象)科研专项"华北日光温室小气候资源高效利用技术研究"(GYHY201306039)

摘　　要：为了系统分析日光温室内外气候特征的关系,向日光温室作物环境调控及小气候预报提供支持,根据冬季日光温室内小气候观测试验资料和附近气象站观测资料,利用BP神经网络方法建立3个模型,分别对3种不同天气状况下石家庄地区日光温室冬季小气候特征进行模拟。结果表明:3个模型气温训练值与实测值的均方根误差(RMSE)都在2℃以内,决定系数都在0.95以上;相对湿度训练值的RMSE都在2个百分点以内,决定系数均高于0.95;接受到的太阳辐射的训练值与实测值的RMSE都在16 W/m2以内,决定系数也均超过0.95。利用此模型得到的气温预测值与实测值的RMSE都在2℃以内,决定系数都在0.9以上;相对湿度预测值与实测值的RMSE都在4个百分点以内,晴天和少云-多云状况下决定系数均高于0.9,寡照状况下的决定系数略低约为0.8;接受到的太阳辐射的预测值与实测值的RMSE都在26 W/m2以内,决定系数均超过0.95。说明所建BP神经网络模型对于不同天气状况下石家庄地区日光温室冬季小气候特征模拟都有较高的精度,可以用于预测。To systematically analyze the relationship of climate characteristics inside and outside the solar greenhouse, and provide support to the solar greenhouse crop environmental regulation and microclimate forecasting, based on the observed data of plastic sunlight greenhouse microclimate and neighboring weather station, by using the method of BP neural network, 3 models were established, to assimilate the microclimate characters under plastic sunlight greenhouse in Shijiazhuang Region during the winter. The results showed that： all of the root mean square error （RMSE） between air temperature trained and measured value from 3 models was no more than 2℃ and the coefficient of determination was more than 0.95 respectively. RMSE between relative humidity trained and measured value was no more than 2 percent points and the coefficient ofdetermination was more than 0.95. RMSE between trained and measured value of solar radiation received was no more than 16 W/m^2 and the coefficient of determination was also more than 0.95. All of the RMSE between air temperature predicted and measured value from the 3 models was no more than 2℃ and the coefficient of determination was more than 0.95 respectively. RMSE between relative humidity predicated and measured value was no more than 4 percent points, and their coefficient of determination was more than 0.9 in sunny or slight cloud-cloudy day, more than that which was 0.8 approximately in sunless day. RMSE between predicted and measured value of solar radiation received was no more than 26 W/m^2 and the coefficient of determination was more than 0.95. The results indicated that 3 BP neural network models had quite precisely for predicting microclimate characteristics under plastic sunlight greenhouse in different weather conditions in Shijiazhuang Region in winter, which could meet the forecast requirements.

关 键 词：日光温室 冬季小气候 BP神经网络 模拟 

分 类 号：S625.5[农业科学—园艺学]